Computational Biology Scope
Use Computational Biology for notes where the biological or chemical object defines the problem: molecule, ligand, protein, pocket, complex, conformer, docking, structure-based modeling, and sequence-level genome objects. Use Molecular Modeling for the molecule/structure/docking-heavy subset. Use AI for method choices that remain meaningful across domains: architectures, learning signals, objectives, generative models, evaluation, and systems.
The same paper can touch both, but the page title should follow the strongest claim:
where can include object, representation, architecture, objective, benchmark, system, or agent workflow.
Naming Rule
| Center | Prefer | Avoid |
|---|---|---|
| Protein, ligand, molecule, pocket, complex, genome sequence | Computational Biology | overbroad biology label |
| Protein sequence, genome region, k-mer, variant-level prediction | Sequence-Based Modeling | Opening broad omics unless needed |
| Molecule identity, ligand role, graph, fingerprint, conformer, property | Molecular and Ligand Modeling | Using Molecular Modeling for every sequence task |
| Protein-ligand pair, target-conditioned activity, affinity, selectivity | Interaction Modeling | Collapsing target-assay context into molecule-only labels |
| Docking, pose, pocket, virtual screening | Structure-Based Modeling | Hiding coordinate assumptions inside generic AI |
| GNN, Transformer, diffusion, flow matching, SSL | AI | Hiding the method under Molecular Modeling |
| Symmetry, coordinate frame, likelihood, loss | Math | Treating equations as paper-specific details |
| Split, metric, leakage, benchmark protocol | Evaluation | Reporting only headline scores |
| Tool-using LLM workflow | Agents | Mixing agent ops into model architecture notes |
Scope Layers
| Layer | Meaning | Examples |
|---|---|---|
| Computational Biology | modeled biological and chemical objects | proteins, ligands, complexes, genomic regions |
| Objects and Entities | vocabulary layer for objects and roles | protein, molecule, ligand, target, pocket, assay |
| Sequence-Based Modeling | sequence-first protein or genome tasks | protein language models, k-mers, genomic regions, variant effect prediction |
| Molecular and Ligand Modeling | molecule-first and ligand-preparation tasks | molecular graphs, fingerprints, conformers, property prediction |
| Interaction Modeling | relation-first tasks | activity, affinity, selectivity, protein-ligand pairs |
| Structure-Based Modeling | coordinate and complex-centered modeling | docking, pose generation, scoring, virtual screening |
Use the broad name when the object family is mixed. Use the narrower name when the note is clearly about molecules, proteins, structures, or genome sequence.
Practical Routes
| Question | Route |
|---|---|
| What is being modeled? | Objects and Entities |
| Is the primary input a biological sequence? | Sequence-based modeling |
| Is the primary object a molecule or ligand state? | Molecular and ligand modeling |
| Is the example a pair, assay row, affinity, or selectivity claim? | Interaction modeling |
| Is it docking, pose, conformer, or screening? | Structure-based modeling, Docking |
| What model family is used? | AI architectures, Learning methods |
| What equation explains the claim? | Math, Formula intake |
| What evidence supports the claim? | Data and evaluation, Evaluation |
| Is the note ready to publish? | AI Computational Biology Math readiness gate |
Examples
| Topic | Primary Home | Secondary Links |
|---|---|---|
| Protein-ligand docking | Docking | Scoring function, Pose RMSD |
| Ligand conformer generation | Molecular and ligand modeling | Conformer, Coordinate modeling contract |
| Equivariant pose prediction | Structure-based modeling | Equivariance, Coordinate prediction |
| Molecular graph pretraining | Molecular and ligand modeling | Self-supervised learning, Graph neural networks |
| Flow-matching molecule generation | Molecular and ligand modeling | Flow matching, Generative models |
Checks
- Does the title name the object or workflow before the model brand?
- Does the note separate domain claims from AI method claims?
- Are conformers, poses, and protein-bound geometries distinguished?
- Are formulas linked to Math notes instead of left as unexplained symbols?
- Are benchmark claims tied to split, metric, baseline, and leakage risk?