Case study
Modality gaps in the bromodomain reader family
Which bromodomain targets lack covalent / degrader / macrocycle chemistry?
The question
Operator-authored prose pending
modality coverage as a whitespace lens: which chemistries have been tried against a family, and which haven't, in a database large enough to make 'never tried' a meaningful signal — operator writes this section as long-form prose, ~150–250 words, footnoted, slow. Scaffold only.
Strategic context
Operator-authored prose pending
BET/BRD family history: pan-BET inhibitors failed broadly; degraders (dBET1, ARV-825 lineage) reopened the space; covalent and macrocycle chemistries have been notably absent — why? — operator writes this section as long-form prose, ~150–250 words, footnoted, slow. Scaffold only.
The Mosaic-derived answer
Operator-authored prose pending
interpret the figure: small-molecule coverage is broad, degraders are concentrated on BRD4 only, and covalent / macrocycle / peptide-like are absent across the whole family — operator writes this section as long-form prose, ~150–250 words, footnoted, slow. Scaffold only.
The graph reasoning, step by step
Operator-authored prose pending
SMILES → heuristic modality classifier → compound_targets join → group by target × modality → absent_modalities = canonical 5 minus present; unclassified compounds reported separately for honesty — operator writes this section as long-form prose, ~150–250 words, footnoted, slow. Scaffold only.
Implications
Operator-authored prose pending
program ideas: covalent BET binders targeting an acquired cysteine; macrocyclic stapled-peptide BD1/BD2 selectivity; BRDT-specific degraders as a contraceptive-adjacent niche — operator writes this section as long-form prose, ~150–250 words, footnoted, slow. Scaffold only.
Limitations & honesty
Operator-authored prose pending
heuristic SMILES classifier is partial; family-level rollup is target-by-target stacked (protein_family is unpopulated upstream); 'absent' = absent from our top-ranked compounds, not literally never made anywhere — operator writes this section as long-form prose, ~150–250 words, footnoted, slow. Scaffold only.