GEPER
Menu

Research

Where Biology Meets Computation.

Research at GEPER focuses on how biological variation can be represented, modelled and interpreted computationally.

These are directions of active work. Nothing on this page is presented as a completed result.

  1. 01Biological data
  2. 02Representation
  3. 03Modelling
  4. 04Interpretation
A genomic signal traced through a computational systemA sequence track with one variant highlighted, connected to gene, protein and population context, passing through layers of computation and converging on a single evidence-weighted result.SEQUENCEVARIANTGENEPROTEINPOPULATIONLAYERSEVIDENCE-WEIGHTED RESULT

signal path

schematic

The research pipeline

How a variant becomes an interpretation.

Four stages, each with its own open questions. The work at every stage constrains what the next one can honestly claim.

  1. 01Biological data

    Variation arrives as sequence.

    A variant is a position where an individual differs from a reference. On its own it means little; its meaning comes from the gene, protein and population around it.

    READS · REFERENCEGENEPROTEIN REGIONPOPULATION

    variant + context

  2. 02Representation

    Biology becomes computable.

    Scattered, inconsistently described observations are normalised into structured objects a model can reason over, with missing information marked as missing.

    OBSERVEDSTRUCTUREDMISSING IS MARKED, NOT GUESSED

    scattered → structured

  3. 03Modelling

    Structure moves through models.

    Representations pass through layers of computation that estimate possible effects. Each estimate is one line of evidence, never a verdict on its own.

    INPUTESTIMATEEACH ESTIMATE IS ONE LINE OF EVIDENCE

    layered estimation

  4. 04Interpretation

    Evidence converges.

    Sequence, protein, population and curated knowledge are weighed together under published frameworks, and the reasoning behind the result is recorded.

    SEQUENCEPROTEINPOPULATIONCURATEDWEIGHTEDEVERY STEP OF THE REASONING IS RECORDED

    weighted convergence

Research directions

Four questions we are working on.

Open directions, stated as questions because that is what they still are.

encoding that preserves context

  • 01

    How should sequences and variants be represented so that computation preserves their biological meaning?

    • Consistent, normalised variant description
    • Sequence and protein context as first-class inputs
    • Representing absence and uncertainty explicitly

  • 02

    How can sequence, structure, expression, conservation and population information be combined?

    • Combining sequence, protein and population evidence
    • Assessing agreement and disagreement between models
    • Understanding what each modality can and cannot tell us

  • 03

    How can computational evidence be connected to potential biological consequences?

    • Evidence aggregation under published interpretation standards
    • Provenance: recording which evidence supported which conclusion
    • Reporting confidence and its limits

  • 04

    Planned

    Can computational models generate useful biological predictions and hypotheses?

    • Hypothesis generation from sequence models
    • Prioritising variants and genes for further study
    • Evaluating predictions against independent evidence

Next

Research becomes useful when it can be built.

These research directions inform the architecture of the GEPER platform and the systems being developed on top of it.

Explore the Platform