AI4SC
Home/Research

RESEARCH AGENDA

Our
research

Biological questions define our computational tasks; testability constrains our models. The AI4SC agenda connects representation, integration, intervention and transcript-level mechanisms across datasets and experimental contexts.

02 / RESEARCH
Single-cell foundation model concept
01 / FOUNDATION MODELS

Single-cell foundation models

Learning transferable representations from large, diverse single-cell collections without mistaking batch, technology or annotation bias for biology.

Core questionHow can cell state, context and hierarchy be represented together?
MethodsPretraining, transfer, interpretability and reliability
Multimodal and spatial intelligence concept
02 / MULTIMODAL & SPATIAL

Multimodal & spatial intelligence

Integrating transcriptomic, epigenomic, proteomic and spatial signals while preserving tissue structure and measurement uncertainty.

Core questionHow can modalities agree without erasing meaningful differences?
MethodsCross-modal alignment, spatial models, graphs, imputation
Perturbation and causal discovery concept
03 / PERTURBATION & CAUSALITY

Perturbation & causal discovery

Learning cell-state transitions from observation and intervention, separating association from mechanisms that can guide experiments.

Core questionWhich regulatory relationships explain and predict change?
MethodsPerturbation models, counterfactuals, trajectories, dynamics
Long-read and isoform intelligence concept
04 / LONG-READ & ISOFORMS

Long-read & isoform intelligence

Resolving full-length transcripts, splicing and isoform regulation to move beyond gene-level averages toward finer molecular mechanisms.

Core questionHow do isoform changes relate to cell identity, state and function?
MethodsTranscript quantification, structure, function, integration

Standards across every direction

  • Biologically valid

    Evaluation must include experimental design, known mechanisms and independent evidence—not only algorithmic metrics.

  • Generalizable

    We specify what transfers across datasets, species, tissues, disease contexts and technologies.

  • Interpretable

    Predictions should become understandable, comparable and testable hypotheses with explicit uncertainty.

  • Reproducible

    Data provenance, versions, code, parameters and evaluation protocols should support independent review.

NEXT / 03Platforms
→