Computational Biology Boundary

Computational Biology in this site means the object and workflow side of modeling biological and chemical systems. It covers molecular, protein, structure, interaction, and sequence-level modeling, but it does not try to survey all of biology. Broad omics, systems biology, transcriptomics, single-cell analysis, and clinical biology stay out of scope unless a concrete research need appears.

The practical unit is:

$$\text{biological object}+\text{measurement context}+\text{computational representation}\to \text{claim}$$

If the object, measurement, representation, or claim is vague, the model result is hard to interpret.

Scope

Area	Use For
Computational Biology Scope	route choice and boundary between domain object, AI method, Math, and Agents
Objects and Entities	protein, molecule, ligand, pocket, complex, assay, sequence, structure
Sequence-based modeling	protein sequence, genome sequence, representation, family or region split
Molecular and ligand modeling	small-molecule identity, standardization, fingerprints, graphs, conformers
Interaction modeling	target-conditioned activity, affinity, selectivity, pair-level claims
Structure-based modeling	protein-ligand geometry, pose, interaction, scoring
Docking	receptor/ligand preparation, pose generation, pose quality, ranking
Data and evaluation	split units, label semantics, leakage, benchmark contracts

Core Questions

• What is the biological or chemical object?
• What is the context: target, pocket, assay, condition, organism, or source?
• What representation is used: string, graph, fingerprint, sequence embedding, coordinate set, or complex graph?
• What split unit supports the claim: scaffold, protein family, complex pair, assay, or time?
• What metric answers the actual task rather than a convenient proxy?

Representation Map

Object	Common Representations	Main Risk
Molecule / ligand	SMILES, molecular graph, fingerprint, conformer ensemble	identity drift from salt, tautomer, protonation, stereo, or standardization choices
Protein	amino-acid sequence, MSA, structure, residue graph, pocket representation	family leakage, residue-index mismatch, missing regions, structure-cleaning artifacts
Protein-ligand complex	pocket-ligand coordinates, interaction graph, distance/contact features	template leakage, ligand-defined pocket leakage, symmetry-incorrect pose comparison
Assay label	endpoint, unit, threshold, censoring, source, target context	mixing incompatible assays or treating weak labels as clean labels
Genome region	sequence window, k-mer counts, variant context, annotation features	coordinate-system mismatch and overbroad omics claims

Claim Template

A public computational biology claim should be reducible to:

$$(\text{object},\text{context},\text{representation},\text{split},\text{metric})\to \text{claim}$$

For example, a docking note should not only say that a model performs well. It should say whether the claim is about pose generation, pose plausibility, ranking, affinity prediction, enrichment, or out-of-family generalization.

Boundaries

Include	Defer
structure-based modeling	broad systems biology
molecular and ligand modeling	transcriptomics and single-cell analysis
sequence-based protein/genome modeling	clinical omics
interaction modeling	pathway biology unless a concrete project needs it
assay-aware labels and splits	private datasets or unpublished results

Related

• Computational Biology
• Computational Biology Scope
• Dataset construction checklist
• Leakage
• Entity relation map
• Molecular modeling concepts
• Protein modeling concepts