From Genomics to AI Biology (Series 2)

Biology Is Not Just Data
Why Context Is the Bottleneck for AI Biology

By Li Lei on July 13, 2026

When I first worked with large-scale genomics data, I remember staring at long tables of significant genes, variants, and enriched pathways.

The results looked impressive.

But the real question was harder:

Which signals were biologically real?

Which were artifacts?

Which were statistically significant but biologically weak?

Which were worth following up on in an experiment?

That experience still shapes how I think about AI biology.

At first glance, genomics looks like a perfect field for artificial intelligence. Biology is producing enormous amounts of data: genome sequences, variants, RNA-seq, ATAC-seq, single-cell profiles, methylomes, proteomics, metabolomics, phenotypes, images, environmental records, and clinical measurements. Over the past decade, deep learning has become increasingly important for genome analysis and interpretation, including regulatory genomics, variant effect prediction, and functional genomics [1,2].

It is tempting to think that biology is simply waiting for better models.

More data.

Bigger models.

More compute.

Better prediction.

But the longer I work at the intersection of genomics, computational biology, and AI, the more I believe this view is incomplete.

Biology is not just a data problem.

Biology is a context problem.

A gene is not just a gene.

A variant is not just a variant.

A cell is not just a cell.

A phenotype is not just a label.

A model prediction is not automatically a discovery.

Biological meaning depends on context: tissue, cell type, developmental stage, genotype, environment, population history, evolutionary constraint, experimental design, assay technology, and validation evidence.

This is why I believe one of the biggest bottlenecks in AI biology is not only model architecture or compute.

It is context.

A prediction is not yet a discovery

Imagine an AI system ranks a gene as a top candidate for drought tolerance in barley or sorghum.

At first, this sounds exciting. The model has scanned a large biological space and found something worth attention. Maybe the gene is near a trait-associated variant. Maybe it is differentially expressed under drought. Maybe it appears in a regulatory network. Maybe the literature suggests a connection to stress response.

But that ranking is not yet a discovery.

It is the beginning of a biological investigation.

A biologist will immediately ask:

Is this gene expressed in the relevant tissue — root, leaf, seed, or reproductive organ?

Is it active at the relevant developmental stage?

Is the response specific to drought, or is it a general stress response?

Is the association driven by population structure or linkage disequilibrium?

Is there chromatin or regulatory evidence?

Is the gene conserved across grasses?

Does perturbing the gene actually change the phenotype?

Would improving drought tolerance create a tradeoff with growth, yield, or reproduction?

These questions are not secondary details.

They are the biology.

This is where AI biology becomes difficult. A model can learn patterns from data, but biological meaning depends on where, when, how, and under what conditions those patterns appear.

In biology, signals are often real, but their meaning is conditional.

That conditionality is exactly what makes biological interpretation hard. It is also what makes AI biology hard.

The same signal can mean different things

A gene highly expressed in leaves under drought stress may be important for water response. But the same gene expressed in roots, seeds, or reproductive tissues may tell a very different story.

A transcription factor may regulate stress response in one developmental stage and growth in another.

A variant may look associated with a trait, but the signal may be driven by population structure, linkage disequilibrium, or environmental stratification.

A chromatin region may be accessible, but that does not automatically mean it is an active enhancer.

A gene may be differentially expressed, but that does not automatically mean it is causal.

A pathway may be enriched, but that does not tell us which mechanism is driving the phenotype.

This is why context is not a detail added after analysis.

Context is part of the biological signal.

If the data does not carry enough context, an AI model may learn shortcuts instead of biology. It may learn species-specific bias, batch effects, tissue composition, sequencing depth, publication bias, or experimental artifacts. It may perform well on a benchmark but fail when applied to a new genotype, tissue, environment, disease cohort, or crop field.

In biology, a prediction is useful only when we understand its boundary.

Biology has many layers of context

When we say “context,” it can sound vague. But in biology, context is not abstract. It has real layers.

I think of biological context as at least seven connected layers.

1. Molecular context

At the molecular level, we ask: what is happening in the genome, epigenome, transcriptome, proteome, or metabolome?

A DNA variant may change a protein sequence. Or it may affect a regulatory element. Or it may do nothing obvious by itself but matter in combination with other variants. A gene expression signal may reflect transcriptional regulation, RNA stability, cell composition, or technical noise.

A sequence is never just a string of letters. It exists in chromatin, in a genome, in a regulatory architecture, and in an evolutionary history.

This is why models such as Enformer are exciting. They show that sequence-based models can improve gene expression prediction by incorporating long-range regulatory information, but they also remind us that regulatory interpretation depends on genomic and cellular context [3].

2. Cellular context

The same gene can behave differently across cell types. This is one of the major lessons from single-cell and single-nucleus biology.

A signal that disappears in bulk RNA-seq may become clear at the cell-type level. A gene may be active only in a rare cell population. A regulatory program may be specific to a transient developmental state. A disease or stress response may not affect all cells equally.

Large efforts such as the Human Cell Atlas were built around this idea: to understand human biology, we need to define cell types by molecular profiles and connect them with location, morphology, and function [4].

AI models that ignore cell type and cell state may average away the biology we actually care about.

3. Tissue context

Biology is spatial and tissue-specific.

A drought response in roots is not the same as a drought response in leaves. A regulatory element active in one tissue may be silent in another. A variant affecting seed development may have little relevance to vegetative growth. A disease marker in blood may not reflect what is happening in the target tissue.

This matters deeply for both plant biology and human biology.

When context is missing, a model may connect the right entities in the wrong place.

4. Organismal context

Genes act inside organisms, not in isolation.

The same variant may have different effects depending on genetic background. The same pathway may behave differently across accessions, cultivars, strains, populations, or species. A phenotype may result from many small effects across the genome rather than one obvious causal gene.

This is something population genetics teaches us very well: variation is structured. History matters. Background matters. Selection matters. What appears as a clean association may be entangled with demography, ancestry, and environment.

AI systems that ignore organismal and population context may confuse correlation with cause.

5. Environmental context

Biology is responsive.

Temperature, drought, light, nutrients, pathogens, microbiome, management practice, diet, treatment, and climate can all change biological meaning. A gene that matters under stress may not matter under control conditions. A genotype that performs well in one environment may fail in another.

In agriculture, this is obvious. A trait is rarely just genetic. It is genotype-by-environment interaction.

In medicine, it is also true. Disease risk, treatment response, and immune state are shaped by environment, history, and exposure.

AI biology must learn not only biological entities, but biological conditions.

6. Evolutionary context

Biology has history.

Some genes are deeply conserved. Others are lineage-specific. Some regulatory elements evolve quickly. Others remain constrained across millions of years. A pathway discovered in Arabidopsis may or may not translate directly to maize, sorghum, rice, or Brachypodium. A signal from mouse may or may not translate to human.

Evolutionary context helps us ask whether a pattern is likely to be functional, conserved, diverged, or species-specific.

Without evolutionary thinking, AI may treat all biological observations as flat examples. But biology is not flat. It is shaped by descent, divergence, constraint, and adaptation.

7. Experimental context

Finally, every biological dataset comes from an experiment.

How was the sample collected?

What protocol was used?

What sequencing platform?

What batch?

What controls?

What replicates?

What normalization?

What quality filters?

What reference genome?

What annotation version?

These details are not technical decoration. They determine what conclusions can be trusted.

This is also why the FAIR principles — making data findable, accessible, interoperable, and reusable — are so important for data-intensive science. They emphasize that data, metadata, tools, and workflows should be reusable by both humans and machines [5].

A beautiful machine learning model trained on poorly annotated, biased, or inconsistent data may produce confident predictions with weak biological value.

In AI biology, metadata is not paperwork.

Metadata is part of intelligence.

Why bigger models are not enough

I am not against bigger models.

Foundation models have already shown remarkable potential in protein biology, genomics, chemistry, and medicine. AlphaFold demonstrated the power of deep learning for protein structure prediction and changed how many scientists think about structural biology [6]. Models such as Enformer show how deep learning can extract regulatory information from DNA sequences at scale [3].

These are real breakthroughs.

But bigger models alone do not solve the context problem.

A model trained only on sequence can learn sequence patterns, but unless tissue-specific activity is represented in the training data or system design, it may not distinguish where a regulatory element is active.

A model trained on expression can learn cell states, but unless treatment, genotype, and sampling conditions are represented, it may not understand why those states appear.

A model trained on literature can learn published knowledge, but it may also inherit publication bias, outdated assumptions, and unsupported claims.

The issue is not whether models are powerful.

They are.

The issue is whether the system around the model carries enough biological context to make the output meaningful.

This is where many AI biology efforts become fragile. They focus heavily on model architecture but underinvest in data provenance, metadata, knowledge representation, benchmark design, biological validation, and feedback loops.

In other words, they build impressive engines but give them incomplete maps.

From prediction to meaning

AI is very good at prediction. But science is not only prediction.

Science also asks:

Why does this happen?

Under what conditions does it happen?

Is the relationship causal?

Can it generalize?

Can we intervene?

What evidence would change our mind?

A prediction becomes biologically valuable only when it helps us move toward understanding, decision, or action.

This is why AI biology needs scientists who understand both models and mechanisms. The model can point us toward a pattern. But the scientist must decide whether the pattern is meaningful, testable, and worth pursuing.

For the drought-tolerance gene example, the model’s ranking may be the first signal. But the biological work begins afterward.

We need to ask whether the gene is expressed in the right tissue, active under the right condition, supported by regulatory evidence, robust across genotypes, conserved across species, and testable through perturbation or validation.

That is how a prediction becomes a hypothesis.

And only through evidence does a hypothesis become discovery.

The danger of context-free AI

Context-free AI can be fast, fluent, and impressive.

It can summarize papers.

It can rank genes.

It can generate hypotheses.

It can produce beautiful explanations.

But if it does not represent biological context, it can also mislead.

It may overgeneralize from one species to another.

It may treat correlation as causation.

It may ignore tissue specificity.

It may miss population structure.

It may combine evidence from incompatible assays.

It may present a weak hypothesis with strong language.

It may sound confident because language models are good at fluency, not because the biology is correct.

This is especially risky because biological AI outputs often look plausible.

A wrong gene ranking may not look obviously wrong.

A weak mechanistic explanation may sound elegant.

A false literature connection may appear reasonable.

A biased model may perform well on an internal benchmark.

The danger is not only that AI can be wrong.

The danger is that AI can be wrong in a biologically plausible way.

That is why context-aware evaluation is essential.

We should not only ask whether a model performs well on average. We should ask where it performs well, where it fails, and why.

Does it generalize across species?

Across tissues?

Across environments?

Across populations?

Across experimental platforms?

Across time?

A model that performs well only in familiar contexts may still be useful. But we must know its boundary.

In biology, knowing the boundary of a model may be as important as knowing its prediction.

Context engineering: making biological meaning computable

If prompt engineering was one of the first popular skills of the AI era, context engineering may become one of the most important skills in AI biology.

By context engineering, I mean the work of making biological context computable.

This is where knowledge graphs, ontologies, metadata standards, multimodal integration, retrieval systems, and validation loops become important.

A knowledge graph can connect genes, variants, traits, tissues, pathways, publications, experiments, species, environmental conditions, and evidence types. Biomedical knowledge graphs are often used to represent biological and clinical concepts as nodes and relationships, making them useful for data integration, machine learning, and reasoning [7].

Graph representation learning extends this idea by learning from network structure and node relationships. In biomedicine, such approaches have been applied across molecular networks, disease relationships, patient data, drug discovery, and other connected biological systems [8].

This does not make the system perfect. But it gives the model a richer map.

Instead of asking AI to reason from isolated data points, we can ask it to reason over structured biological relationships.

This is especially important for AI agents in science.

If an agent is expected to answer biological questions, design analyses, prioritize candidates, or propose experiments, it needs more than a language model. It needs access to reliable data, structured knowledge, metadata, literature, tools, evaluation criteria, and human feedback.

An AI agent without biological context is like a brilliant student with no lab notebook, no experimental design, and no memory of what was actually measured.

It may be clever, but it is not yet a scientist.

What context-aware AI biology might look like

A context-aware AI biology system would not simply take a gene list and return a polished interpretation.

It would ask better questions.

What species is this from?

Which tissue?

Which developmental stage?

Which treatment?

Which genotype?

Which assay?

Which controls?

Which reference annotation?

Which evidence is direct and which is inferred?

Which claims come from literature and which come from data?

Which results are robust across datasets?

Which hypotheses are testable?

It would not only produce an answer.

It would show the evidence path.

It would connect a prediction to the data, the metadata, the prior knowledge, the uncertainty, and the next experiment.

This is the kind of AI biology I find exciting.

Not AI that simply gives us more answers.

AI that helps us ask better questions.

What biologists should contribute

In the AI era, biologists should not underestimate their role.

It is easy to feel that machine learning experts are driving the future and biologists are only data providers. But that would be a mistake.

Biologists understand what the question means.

Biologists understand whether the data is appropriate.

Biologists understand what the model might be missing.

Biologists understand what evidence is convincing.

Biologists understand whether a prediction is experimentally actionable.

This expertise is not secondary.

It is central.

The future of AI biology needs biologists who can define meaningful problems, identify relevant context, design better evaluation, and interpret model outputs with scientific discipline.

This does not mean every biologist must become a machine learning engineer. But more biologists should learn enough AI to participate in system design, not only tool usage.

They should be able to say:

This dataset is not enough.

This benchmark is misleading.

This prediction is interesting but not actionable.

This context is missing.

This result needs validation.

This hypothesis is worth testing.

That is scientific leadership in the AI era.

My own lesson from genomics

Genomics taught me humility.

When I first learned to analyze large biological datasets, I was fascinated by the power of scale. Suddenly, we could look across the genome, across populations, across tissues, across conditions. We could move from one gene at a time to thousands of genes at once.

But scale also made the problem harder.

More data did not automatically create more understanding. Sometimes it created more ambiguity. More correlations. More candidates. More possible explanations.

The real work was to connect signals back to biology.

Which signal is robust?

Which one is confounded?

Which one generalizes?

Which one has mechanism?

Which one can be validated?

Which one matters?

AI biology is entering the same stage.

We are surrounded by new models, new tools, new predictions, and new promises. The opportunity is real. But the central question remains deeply scientific:

What does it mean?

Biology is data with meaning

I believe AI will transform biology. It will help us read genomes, design proteins, interpret cells, mine literature, integrate knowledge, generate hypotheses, and accelerate discovery.

But the most important progress will not come from treating biology as a giant spreadsheet.

It will come from building systems that understand relationships, constraints, mechanisms, uncertainty, and context.

Biology is not just data.

Biology is data with history.

Data with structure.

Data with environment.

Data with measurement noise.

Data with evolutionary memory.

Data with experimental limits.

Data with meaning.

The future of AI biology will depend on how well we teach machines to work with that meaning.

And that work cannot be done by AI alone.

It requires biologists who understand data, computational scientists who understand models, engineers who understand systems, and teams that respect the complexity of life.

In the first essay of this series, I wrote that AI will not replace biologists, but biologists who understand AI systems will replace those who only use AI tools.

In this second essay, I want to go one step deeper:

Biologists who understand context will be essential to the future of AI biology.

Because in biology, context is not background.

Context is the bottleneck.

And perhaps, context is also the key.

If we want AI to become truly useful for biology, the question is not only how large the model is.

The question is:

What context does the system need in order to make a biological answer trustworthy?

That is the conversation I hope more biologists will join.

References and further reading

[1] Zou, J., Huss, M., Abid, A., Mohammadi, P., Torkamani, A. & Telenti, A. A primer on deep learning in genomics. Nature Genetics 51, 12–18 (2019). doi:10.1038/s41588-018-0295-5.

[2] Eraslan, G., Avsec, Ž., Gagneur, J. & Theis, F. J. Deep learning: new computational modelling techniques for genomics. Nature Reviews Genetics 20, 389–403 (2019). doi:10.1038/s41576-019-0122-6.

[3] Avsec, Ž. et al. Effective gene expression prediction from sequence by integrating long-range interactions. Nature Methods 18, 1196–1203 (2021). doi:10.1038/s41592-021-01252-x.

[4] Regev, A. et al. The Human Cell Atlas. eLife 6, e27041 (2017). doi:10.7554/eLife.27041.

[5] Wilkinson, M. D. et al. The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data 3, 160018 (2016). doi:10.1038/sdata.2016.18.

[6] Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). doi:10.1038/s41586-021-03819-2.

[7] Nicholson, D. N. & Greene, C. S. Constructing knowledge graphs and their biomedical applications. Computational and Structural Biotechnology Journal 18, 1414–1428 (2020). doi:10.1016/j.csbj.2020.05.017.

[8] Li, M. M., Huang, K. & Zitnik, M. Graph representation learning in biomedicine and healthcare. Nature Biomedical Engineering 6, 1353–1369 (2022). doi:10.1038/s41551-022-00942-x.

From Genomics to AI Biology (Series 1)

 AI Will Not Replace Biologists. But Biologists Who Understand AI Systems Will Replace Those Who Only Use AI Tools

By Li Lei on June 21, 2026

When I first moved from genomics and computational biology into AI biology, I thought the biggest challenge would be learning new models.

Large language models. Graph neural networks. Foundation models. AI agents. Embedding spaces. Retrieval-augmented generation. Model evaluation. MLOps.

All of these were important. But over time, I realized something deeper:

The real challenge is not simply learning how to use AI tools.

The real challenge is learning how to think in AI systems.

This distinction matters because biology is entering a new era. AI is no longer just a convenient assistant for writing code, summarizing papers, or generating figures. It is beginning to shape how we organize knowledge, design experiments, prioritize hypotheses, analyze multimodal data, and make decisions in discovery pipelines.

We have already seen powerful examples. AlphaFold changed how many scientists think about protein structure prediction [1]. Protein language models showed that large-scale learning from sequences can capture biological structure and function [2]. Deep learning models such as Enformer demonstrated that sequence-based models can improve gene expression prediction by modeling long-range regulatory information [3]. More broadly, deep learning has become an important part of modern genomics and genome interpretation [4,5].

But these advances also reveal something important: the future of AI biology is not only about bigger models. It is about building better scientific systems around those models.

In this new era, the biologists who thrive will not necessarily be the ones who use the most AI tools. They will be the ones who understand how AI systems work, where they fail, how to evaluate them, and how to connect them to real biological questions.

That is why I believe:

AI will not replace biologists. But biologists who understand AI systems will replace those who only use AI tools.

From using tools to understanding systems

Many scientists today are already using AI. They ask ChatGPT to polish manuscripts, write Python scripts, explain error messages, summarize literature, or draft emails. These uses are helpful. They save time. They reduce friction. They make technical work more accessible.

But using AI as a tool is only the first layer.

A tool user asks:

“Can AI help me finish this task faster?”

A systems thinker asks:

“What is the input? What is the output? What knowledge does the model have? What knowledge is missing? How do I evaluate whether the answer is correct? How does this fit into a larger scientific workflow?”

This difference may sound small, but it changes everything.

For example, if I ask an AI model to summarize papers about drought tolerance in barley, I may receive a fluent answer. But a fluent answer is not necessarily a reliable answer. A biologist who only uses the tool may accept the summary too quickly. A biologist who understands AI systems will ask: Which papers were retrieved? Are they current? Are the claims supported by experimental evidence? Are the genes discussed in the right tissue, developmental stage, and environmental context? Are we mixing evidence from Arabidopsis, rice, sorghum, and barley as if they were interchangeable?

In biology, context is not decoration. Context is the science.

A gene is not simply “important.” It is important in a genotype, tissue, cell type, developmental stage, environment, and evolutionary history. A variant is not simply “associated.” It has allele frequency, linkage disequilibrium, population structure, effect size, uncertainty, and biological plausibility. A regulatory element is not simply “predicted.” It has chromatin accessibility, transcription factor binding, conservation, activity, target gene ambiguity, and experimental validation limits.

This is why biological AI requires more than prompting. It requires system-level thinking.

Biology is not just data

One common misunderstanding in AI biology is the idea that biology is simply a data problem.

More data, bigger model, better prediction.

Sometimes that is true. Often, it is incomplete.

Biological data is noisy, biased, incomplete, heterogeneous, and deeply contextual. Different data types capture different layers of life: genome sequence, chromatin accessibility, gene expression, methylation, protein structure, metabolites, phenotypes, environmental variables, clinical outcomes, and evolutionary constraints. Each layer has its own measurement errors, assumptions, and missingness.

A machine learning model can find patterns. But not every pattern is meaningful. Not every correlation is causal. Not every prediction is actionable.

This is where trained biologists remain essential.

Biologists understand experimental design. They understand confounding. They know that a beautiful heatmap can hide a batch effect. They know that a significant association can be driven by population structure. They know that a gene expression signal may reflect cell type composition rather than regulation. They know that a model trained on one species, tissue, or condition may not generalize to another.

AI can accelerate discovery, but it does not automatically understand what makes a biological conclusion trustworthy.

That judgment still comes from scientists.

The future belongs to scientists who can combine biological judgment with AI system design.

The next skill is not just coding

For the past decade, many biologists were told: “Learn to code.”

That advice was useful. Coding opened the door to bioinformatics, genomics, data analysis, and reproducible research. It allowed biologists to work directly with large datasets rather than relying entirely on others.

But in the AI era, coding alone is no longer enough.

The next skill is understanding how biological knowledge becomes computable.

This includes questions such as:

How do we represent biological entities and relationships?

How do we connect genes, variants, traits, pathways, tissues, environments, publications, and experimental evidence?

How do we integrate structured databases with unstructured literature?

How do we build workflows where AI agents can retrieve, reason, analyze, and report?

How do we evaluate whether an AI-generated hypothesis is biologically meaningful?

How do we prevent models from producing confident but unsupported conclusions?

These are not just computer science questions. They are scientific questions.

A good AI system for biology is not just a model. It is a carefully designed connection between data, knowledge, algorithms, evaluation, and human decision-making.

That is why I believe knowledge representation will become one of the most important skills in AI biology.

Biomedical knowledge graphs already show why representation matters. They provide a way to connect entities such as genes, proteins, diseases, drugs, phenotypes, pathways, and publications into structured relationships that both humans and machines can query and reason over [6]. Graph representation learning further extends this idea by learning from the topology and semantics of biological and biomedical networks [7].

In genomics, we often start with sequences. But discovery rarely ends with sequence alone. We need to connect sequence variation to gene regulation, gene regulation to cellular function, cellular function to phenotype, and phenotype to environment or disease. This chain is complex. It is full of uncertainty. But it is also where the real biological meaning lives.

AI systems that ignore this complexity may generate answers. AI systems that model this complexity may generate insight.

The danger of becoming only an AI consumer

There is a risk in the current AI wave: scientists may become passive consumers of AI outputs.

The model suggests a candidate gene.

The model ranks a variant.

The model proposes a pathway.

The model writes the interpretation.

If we are not careful, scientists may slowly lose the habit of questioning the reasoning behind the output.

That would be dangerous.

Science advances through skepticism. We ask why. We ask how. We ask what evidence supports the claim. We ask whether there is another explanation. We ask what experiment could prove us wrong.

AI should not weaken this habit. It should make it stronger.

A biologist who understands AI systems does not blindly trust the model. But she/he/they also does not reject it out of fear. Instead, she treats AI as a powerful but imperfect collaborator.

she/he/they asks:

What data was this model trained on?

What assumptions are built into the system?

What is the failure mode?

What kind of uncertainty is being hidden?

What evidence would increase my confidence?

What experiment should come next?

This is the mindset we need.

Not AI worship.

Not AI fear.

AI literacy with scientific discipline.

What should biologists learn now?

Not all biologists need to become machine learning engineers. But I do think more biologists need to understand the architecture of AI-enabled discovery.

At minimum, future-ready biologists should understand five things.

First, they should understand data. Not only how to download it, but how it was generated, normalized, biased, and limited.

Second, they should understand representation. In biology, how we represent a problem often determines what the model can learn. A sequence, a graph, a table, an image, a time series, and a knowledge graph all expose different aspects of the same biological system.

Third, they should understand models. They do not need to derive every equation, but they should know what different models are good at, what they assume, and when they are likely to fail.

Fourth, they should understand evaluation. In AI biology, a high benchmark score is not the same as biological usefulness. We need to evaluate models based on generalization, interpretability, robustness, experimental relevance, and decision value. Recent discussions of large language models in scientific discovery also emphasize that these systems should be integrated into scientific workflows with clear human goals and clear evaluation metrics [8].

Fifth, they should understand workflows. AI is most powerful when embedded into real scientific workflows: literature mining, data integration, hypothesis generation, prioritization, experiment design, and feedback from validation.

This is the shift from using AI tools to building AI-assisted scientific systems.

A personal transition

My own path into AI biology did not start from computer science. It started from population genetics, evolutionary biology, and genomics.

Population genetics trained me to think about variation, structure, uncertainty, history, and selection. Genomics trained me to work with large-scale biological data. Bioinformatics trained me to build pipelines and extract signals from complexity. AI is now teaching me to think about representation, reasoning, automation, and decision systems.

Each stage did not replace the previous one. It expanded it.

This is why I do not see AI as a departure from biology. I see it as a new language for asking biological questions.

But learning this language requires humility.

We need to admit that many AI methods are unfamiliar. We need to learn new concepts. We need to collaborate with engineers, data scientists, and machine learning experts. But we also need to remember that biological insight is not outdated. It is more important than ever.

The scientist of the future will not be defined by one discipline. She/he/they will be able to move between biology, computation, data infrastructure, AI models, and real-world decisions.

She/he/they will not simply ask, “What can this tool do?”

She/he/they will ask, “What kind of scientific system are we building?”

The future biologist

The future biologist will still care about genes, cells, organisms, evolution, disease, crops, ecosystems, and patients.

But she/he/they will also understand embeddings, knowledge graphs, agents, multimodal data, model evaluation, and feedback loops.

She/he/they will know how to ask good biological questions and how to design AI systems that make those questions computable.

She/he/they will be skeptical but not afraid.

Technical but not narrow.

Biological but not limited by traditional boundaries.

Curious enough to learn new tools, and wise enough not to be ruled by them.

AI will change biology. There is no doubt about that.

But the deepest transformation will not come from replacing scientists. It will come from changing what scientists are capable of doing.

The most valuable biologists in the AI era will not be those who simply use AI to work faster.

They will be those who understand enough biology to ask meaningful questions, enough AI to build powerful systems, and enough scientific judgment to know when the answer is real.

References and further reading

[1] Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). doi:10.1038/s41586-021-03819-2.

[2] Rives, A. et al. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences. Proceedings of the National Academy of Sciences 118, e2016239118 (2021). doi:10.1073/pnas.2016239118.

[3] Avsec, Ž. et al. Effective gene expression prediction from sequence by integrating long-range interactions. Nature Methods 18, 1196–1203 (2021). doi:10.1038/s41592-021-01252-x.

[4] Zou, J., Huss, M., Abid, A., Mohammadi, P., Torkamani, A. & Telenti, A. A primer on deep learning in genomics. Nature Genetics 51, 12–18 (2019). doi:10.1038/s41588-018-0295-5.

[5] Eraslan, G., Avsec, Ž., Gagneur, J. & Theis, F. J. Deep learning: new computational modeling techniques for genomics. Nature Reviews Genetics 20, 389–403 (2019). doi:10.1038/s41576-019-0122-6.

[6] Nicholson, D. N. & Greene, C. S. Constructing knowledge graphs and their biomedical applications. Computational and Structural Biotechnology Journal 18, 1414–1428 (2020). doi:10.1016/j.csbj.2020.05.017.

[7] Li, M. M., Huang, K. & Zitnik, M. Graph representation learning in biomedicine and healthcare. Nature Biomedical Engineering 6, 1353–1369 (2022). doi:10.1038/s41551-022-00942-x.

[8] Zhang, Y. et al. Exploring the role of large language models in the scientific method: from hypothesis to discovery. npj Artificial Intelligence 1, Article 14 (2025). doi:10.1038/s44387-025-00019-5.

[9] Bommasani, R. et al. On the opportunities and risks of foundation models. arXiv:2108.07258 (2021).

Icebreaker Speech for Toastmaster club: “Who Am I?”

 

Icebreaker Speech for Toastmasters Club: “Who Am I?”

by Li Lei in RTP, NC, in June, 2026

Good afternoon, everyone!

Today is my icebreaker speech at ToastWhisper Club here at Syngenta. Since many of you already know me from work, I want to begin with a simple question:

Who do you think I am?

Maybe some of you would say, “Li is an AI scientist.”
Some may say, “She works in computational biology.”
Some may say, “She is always talking about data, genes, models, and pipelines.”
And some of you may say, “She is the new colleague who is still trying to figure out where everything is in this building.”

All of those are true.

But today, I want to tell you a little bit about the person behind the job title.

I was born and raised in Zhaotong, a small city in Yunnan Province in southwest China. Yunnan is famous for its mountains, flowers, ethnic diversity, and beautiful landscapes. Zhaotong is not a big city, but it shaped me deeply. It gave me curiosity, imagination, and maybe also a little bit of stubbornness.

When I was a little girl, I had a very clear dream: I wanted to become a mathematician.

Not because I fully understood what mathematicians did every day. I did not imagine myself standing in front of a blackboard writing equations for the rest of my life. The real reason was simpler — and a little rebellious.

I heard people say, “Girls are not good at math, especially when they get older.”

I remember thinking, “Really? Who decided that?”

So naturally, I wanted to prove them wrong.

At that time, I learned about Emmy Noether, a brilliant German mathematician. She became one of my role models. She lived in a time when women faced many barriers in academia, but her work changed modern mathematics and physics. To me, she represented intelligence, courage, and quiet strength.

So I studied math very seriously. I loved the beauty of numbers and logic. Math felt like a world where every problem had a hidden door, and if you were patient enough, you could find the key.

My math grades were excellent, and I later won a silver medal in a national mathematics competition. For a young girl from a small city, that was a big encouragement. It made me believe that many limits people place on us are not always real. Sometimes they are just walls built by other people’s assumptions.

But math was not my only dream.

I also wanted to become a poet.

That was a very different dream. Math gave me structure. Poetry gave me freedom. Math helped me understand the world through logic. Poetry helped me feel the world through language.

But when I was young, I also heard people say, “Poets are usually not very happy.” Many famous poets had difficult lives. Some were lonely, depressed, or died young.

So I thought, “Well… maybe being a professional poet is a little dangerous.”

I decided poetry could stay with me as a hobby — a private garden in my heart — but maybe not as my profession.

Looking back, I find this funny. As a child, I gave up being a poet because I thought it was emotionally risky. Then I became a scientist — which is also emotionally risky, just in a different way.

In science, experiments fail. Code breaks. Models do not converge. Papers get rejected. Funding is uncertain. And sometimes, after months of analysis, the data simply tells you, “No.”

So maybe scientists and poets are not so different. Both are searching for patterns. Both are trying to express something true. One uses equations and data; the other uses images and words.

Today, I am an AI scientist, a computational biologist, and a population geneticist. That may sound far away from the little girl who wanted to become a mathematician and poet. But I feel those dreams are still inside me.

As an AI scientist, I still use mathematical thinking. I work with models, algorithms, data, and uncertainty. As a computational biologist, I study life through patterns hidden in genomes and biological systems. As a population geneticist, I think about evolution — how life changes over time, how diversity emerges, and how history leaves traces in DNA.

In some way, I did not abandon math. I followed math into biology.

And I did not abandon poetry either. I still love language, stories, and the beauty of expression. That is one reason I joined Toastmasters. I want to become not only a better scientist, but also a better communicator. I want to learn how to tell stories, how to speak clearly, and how to connect ideas with people.

Because in science, having good ideas is important. But being able to communicate those ideas is equally important. A good idea that no one understands is like a beautiful poem locked in a drawer.

So, who am I?

I am a girl from Zhaotong who once dreamed of becoming a mathematician.
I am someone who still carries poetry quietly inside her.
I am a scientist who studies life through data.
I am a colleague, a learner, and now, a Toastmaster beginning a new journey.

And perhaps, like all of us, I am still becoming.

Thank you.

In Memory of the Great Teachers I Have Met in My Life (IV)

By Li Lei April 13, 2026, in Itaca, NY

This is directly translated from the Chinese version of Blog (https://mp.weixin.qq.com/s/_N62P_8Cbh2fsgiIp1Dyjw)!

Epigraph

A teacher’s guidance is like the spring breeze; a teacher’s grace runs as deep as the sea.

Han Yu once wrote in On Teachers: “A teacher is one who transmits the Way, imparts knowledge, and resolves doubts.” When I first encountered that line, I understood it only in the most formal sense. It seemed to belong to classrooms, lectures, books, examinations. Only later, with time and distance, did I begin to understand how much larger the word teacher really is.

In life, the rarest gift is not simply to meet someone who teaches you facts or skills, but to meet those who quietly shape the deeper architecture of your mind. They steady you when you are still uncertain. They lend you their vision before you have fully formed your own. They may not always speak in maxims, and they may not even look, at first glance, like the solemn figures we imagine teachers to be. But years later, when you turn around and look back, you realize that much of what now feels most deeply yours—your habits of thought, your sense of proportion, your way of seeing the world—was gently marked by their presence.

I have been thinking of the mentors who left such marks on my life. They taught me far more than what could be contained in a syllabus or a degree. They enlarged my world, sharpened my mind, and, in quiet ways, helped form my character.

This essay is about the last of those great teachers in my life, and the one who shaped my professional path most profoundly: my postdoctoral advisor, Peter Morrell.

If I had to name the person who truly taught me how to do scholarship—how to think, how to build an intellectual life with rigor and soul—it would probably be Peter.

It was eleven years ago, in the fall. There was already a touch of coolness in the air. At the time, I was looking for a postdoctoral position in a lab that worked more directly on crops and on questions with a stronger applied dimension. By chance, I heard from a friend of mine that Peter might be recruiting a postdoc, though no one seemed sure whether the position was still open. I looked through his lab website, liked what I saw, and decided to write to him, thinking it was worth trying.

He replied almost immediately and soon arranged an interview.

I still remember that interview with unusual clarity. It remains in my mind like a stretch of autumn light—clear, unhurried, quietly bright.

I had expected the usual format: a formal presentation, a polished summary of my previous work, a series of serious questions. But Peter did not interview me that way at all. Instead, he simply talked with me. We began by discussing major figures in evolutionary biology—Masatoshi Nei, Chung-I Wu, Wen-Hsiung Li—and from there wandered into the intellectual lineages behind them, the ways ideas move across generations and shape a field over time. Later he told me about his own projects, the questions he was thinking about, and even, almost casually, what life in the Twin Cities might be like if I moved there.

Before the conversation ended, he only suggested that I speak with some of the students in the lab, so I could get a feel for the department, the resources, and the atmosphere.

A week later, he sent me an offer. Not long after that, he helped me begin the process of applying for a work visa.

I remember being struck by how direct and efficient he was. There was no performative formality, no dragging things out, no unnecessary complication. His style felt like the season itself—clean, brisk, and unexpectedly generous.

When I finally arrived at the lab, I found myself thinking something embarrassingly unacademic: he was very handsome.

Not in a flashy way, but with an ease and brightness that gave him, to my eyes, something of the aura of an old movie star. At the same time, the lab itself was surprisingly modest. Because most of the work was computational, there was not much equipment, and whenever wet-lab work entered the picture, everything felt a little makeshift. Peter, too, did not fit my old idea of what a mentor should be. He sometimes arrived a few minutes late to meetings. Occasionally he brought his dog—or even his baby—into the lab. In the middle of a scientific discussion, he might suddenly drift into a story about everyday life, then drift back again as if the shift required no explanation at all.

At first, I did not quite know what to make of him.

The mentors I had known before him had all been serious, formal, and controlled—people who wore the role of “advisor” with unmistakable authority. Peter was different. He was relaxed, informal, quick to joke, and unconcerned with appearing impressive in the conventional way. He overturned my expectations almost immediately. Yet there was also something unmistakably old-fashioned about him, in the best sense. If he saw someone wearing ripped jeans to play basketball, he would sincerely wonder whether the person was struggling financially, and might genuinely consider offering money to buy them food.

At the time, I did not yet understand that someone could look so loose on the surface and yet be so deeply grounded underneath. He never performed authority, but he possessed it. He never seemed rigid, but he always had measure. He was like a tree in the wind: the branches moved freely, but the roots were deep.

As I got to know him better, I came to see that Peter’s greatest influence on me did not lie in any single lesson. It lay in the way he changed my understanding of scholarship itself.

The first thing he taught me was how a real project begins.

In his view, research was never supposed to start with “let’s gather a mountain of data and see what we can make of it.” He had little patience for work that relied on accumulating material first and finding a story afterward, or for papers built on descriptive analysis alone, without a real question at their center. He had a particular disdain for shallow projects that could be published simply because they were data-rich, even if they lacked conceptual depth.

For him, scholarship began elsewhere. It began with reading deeply. With learning the landscape of the literature until you could see where the field had already gone, where there was already light, and where something remained obscure. Only then could you ask a question worth asking. Only then could you build a defensible hypothesis. And only then could you design analyses and experiments in a way that allowed evidence to gather naturally around that question.

That way of working reshaped me.

What made his style especially distinctive was that he rarely asked us to give the standard kind of oral presentation about our projects in lab meeting. Instead, when a project was just beginning, he often had us start by drafting the outline of the eventual paper. Not after the results were in. Not after the data had accumulated. At the very start.

What is this paper actually trying to answer?
What is the central thread?
How should the story unfold?
What literature is missing?
What analyses need to be done?
What role should each figure play?

The outline came first. The storyboard came first. Everything else—the analyses, the experiments, the figures—grew from that underlying structure.

That was when I first understood, in a deep way, that good research is not only about generating results. It is about shaping form. It is about building an internal logic strong enough to hold everything together. Research is not a pile of materials; it is an organism. It must grow according to a pattern.

At first, I worried that because we did not give many formal presentations in the lab, my speaking skills might weaken. Later I realized Peter was training us in something more demanding than performance.

He created a small course designed for people outside bioinformatics, and instead of teaching it entirely himself, he had the members of the lab take turns. We were responsible for preparing lectures, designing handouts, creating assignments, and explaining difficult bioinformatics concepts to people from very different backgrounds. The course became so well known that even people from another campus began attending. Often, dozens of people would show up.

Looking back, I now see how brilliant that was.

It is one thing to present your own work to people already fluent in your field. It is another thing entirely to take something complex, abstract, and intimidating, and make it understandable to those who do not already speak the language. That kind of teaching forces you to do more than talk well. It forces you to understand.

Peter was, in this way, teaching us not merely how to present, but how to communicate. Not merely how to appear knowledgeable, but how to make knowledge available to others. There is a difference, and he understood it.

His weekly lab meetings shaped me just as deeply.

They were not project updates. They were paper discussions. And he had one strict rule: no slides.

At first, I found this baffling. In my previous academic environments, a lab meeting almost automatically meant PowerPoint—figures, summaries, clean takeaways, a carefully managed structure. But Peter insisted on stripping all of that away. You brought the paper. You read it. You explained it. You argued with it.

It took me time to understand why.

Without slides, there was nothing to hide behind. No design polish, no visual scaffolding, no way to rely on style instead of substance. You had to read carefully. You had to think carefully. And because nothing mediated your attention, the discussion went more directly to the paper itself.

The range of papers he chose was astonishingly broad. We read foundational classics from the early decades of the field, first papers by major thinkers, brand-new bioRxiv preprints, beautifully written papers, badly written papers, high-level research articles, and papers about gender bias, diversity, and undergraduate education.

At first glance, this might have looked eclectic. In reality, it was a method.

He was training our taste.

After each paper, he would ask us: What works here? What is genuinely original? What is elegant? What is weak? What could have been done better? Over time, this cultivated something invaluable: the ability to read without submission. Not to be overawed by famous names. Not to confuse prestige with quality. To admire real brilliance where it existed, but also to question it when necessary. To see that even imperfect papers can contain sparks of insight, and even celebrated work can have blind spots.

This habit of reading critically—and reading with taste—has stayed with me ever since.

But those discussions shaped more than my scientific judgment. They also sharpened my sensitivity to the broader ethical and social atmosphere of academia. Through them, I became more alert to the ways bias—gendered, racial, institutional—moves through academic life. I began to understand that scholarship is not only about findings and methods. It is also about the climate in which people are trained, heard, dismissed, encouraged, or ignored.

There is another thing about Peter that left a lasting impression on me: he cared deeply about reputation.

I do not mean reputation in the shallow sense—not status, not applause, not ambition for prestige. I mean something older, almost classical. Something closer to the Chinese ideal of the scholar who guards his name because he guards the integrity behind it. Peter had that quality. He treated his writing as an extension of his moral seriousness.

He was extraordinarily demanding of his own papers. Manuscripts that already seemed, to anyone else, quite strong could suddenly be torn down and rebuilt from the beginning. He would rethink the structure, revise the logic, sharpen the language, and begin again if necessary. Even after a paper had been accepted, he might continue polishing it—revising, smoothing, refining—until it felt fully worthy of his name.

He approached reviewers’ comments with the same seriousness. No matter how minor or sharp the criticism, he responded with care and sincerity. He answered each point thoughtfully, trying to understand what the reviewer was really asking, and how the paper might actually become better because of it. There was nothing performative in this. It was not strategy. It was character.

That affected me deeply.

He taught me that scholarship is not simply the production of results. It is also the outward expression of a person’s standards. The way someone handles their own writing says something about how they handle their own conscience. Real rigor is not about impressing other people. It comes from refusing to be careless with one’s own name.

Peter also brought me back to reading.

I had always admired the breadth of his knowledge, but over time I realized that it did not come only from professional expertise. He was simply a true reader. He read science, of course, but also history, mathematics for non-specialists, social thought, and all kinds of books that had nothing immediate to do with work.

Under his influence, I slowly recovered a reading life I had lost. Reading stopped being merely instrumental—something done to solve a problem or complete a task—and became again what it had once been for me: a private source of nourishment and delight.

There is a special kind of relief in opening a book at night that has nothing to do with productivity. It is like opening a window in a life that has grown too narrow. Air enters. Space returns. The mind remembers it was not made to live by utility alone.

What looks irrelevant to scholarship often sustains it most deeply. A mind cannot become spacious if it lives only inside the limits of a profession.

Peter also influenced me in many smaller, subtler ways.

He taught me things about networking, about storytelling, about how to express ideas with both clarity and appeal. More importantly, he had a rare gift for recognizing what was singular in other people.

He seemed genuinely delighted by the fact that I could almost recite the academic lineages of major scholars from memory. He would often ask me, unexpectedly, “What’s this person’s background?” He also appreciated my eye for figures and data visualization, and often encouraged others in the lab to show me their figures and ask what I thought.

To be seen like that by a mentor matters more than simple praise. It is a form of recognition that gives shape to your emerging self. It tells you that what others might overlook in you may in fact be part of your gift.

To be seen that way while you are still young is no small fortune. Many people live for decades without ever fully learning where their own particular light resides.

There is an old Chinese saying: “One day as teacher, a lifetime as father.” Though I left Peter’s lab more than ten years ago, we have remained in touch.

Even now, we still exchange messages from time to time. If he comes across a paper he finds interesting, he may invite me into the discussion. When he writes a review, he sometimes sends it to me first and asks what I think. And later, when I ran into confusion in my own academic life—or into people and situations so absurd they could only be described with a weary laugh—he was still someone I could turn to.

That continuity reminds me of something I once heard about his own academic lineage: even accomplished scholars, long after becoming senior figures themselves, still return to their mentors for guidance. That, to me, is one of the truest signs of a real teacher-student bond. It does not end with graduation. It does not dissolve when a contract ends or a move is made. It changes form, but it remains.

It is like a river that remembers its source even after it reaches the sea. Like a tree whose branches have long spread outward, but whose roots remain sunk deep in the same earth.

Some people teach you for a season. Others become part of the internal conversation of your life.

There is one final thing about Peter that I have always admired: his openness to the new.

From the early days of Sanger sequencing and writing software in Java, to Perl, Python, and now AI, he has always remained willing to learn, adapt, and engage with what is emerging. This has never struck me as trend-chasing. It comes from something deeper—a scholar’s refusal to let knowledge harden into habit.

Some people grow older and gradually become closed rooms, sealed against new air. Others become more open with time, not less. Peter belongs to the latter kind.

He taught me that the best scholars are not those who remain permanently perched atop what they already know, but those who keep the courage to walk toward what they do not yet understand. Scholarship is not a territory one arrives at once and for all. It is a life of repeated departures.

Looking back now, I feel more and more strongly that what Peter gave me was never just a set of research practices or a handful of useful professional lessons.

He gave me a frame for scholarship.
A standard for intellectual honesty.
A way of carrying seriousness without stiffness, rigor without cruelty, openness without loss of depth.

He taught me that scholarship can be exacting without becoming sterile. That thought can be sharp without becoming harsh. That one can hold very high standards and still remain warm, humorous, and deeply human.

Perhaps that is what a great teacher really gives us—not only knowledge, but a fuller way of being.

When I think of Peter now, what comes to mind is not one dramatic moment, but an accumulation of light. A conversation. A comment on a draft. A way of reading. A way of revising. A habit of asking better questions. A standard quietly held.

Human encounters are mysterious that way. Some people pass through our lives and disappear into the crowd. Others remain, like a fine spring rain that seemed almost silent when it fell, but whose work becomes visible later, when all the grass has grown.

I offer this essay in gratitude to my mentor, Peter Morrell—
for his teaching, his example, and for all that continues, even now, to cast light across the years.