SMA Research Platform

The DOK7-MuSK-Agrin-LRP4 ternary complex is the Top-10 #1 flagship NMJ-rescue mechanism in the platform. Per Phase 2 of the breakthrough plan, AF3 + Boltz-2 + Chai-1 + NeuralPLexer-3 consensus on the complex (not just pairs) is the platform's technical moat. We commit to a falsifiable, calendar-bound prediction settled entirely on internal compute — no wet-lab dependency.

By 2026-08-04 (90 days from plan lock), the platform's 4-model cross-consensus engine (Boltz-2 + AF3 + Chai-1 + NeuralPLexer-3) will produce a CONFIRMED status (>= 3 of 4 models, all iPTM >= 0.65, spread < 0.30) for AT LEAST ONE of these canonical NMJ sub-pairs: DOK7 x MuSK_kinase, AGRN_LG3 x LRP4_beta12, or MuSK_frizzled x AGRN.

Each sub-pair is queued through the Spark #2 AF3 lane, Dell Boltz-2 lane, NIM Chai-1 lane, and Modal NeuralPLexer-3 lane (after weights download completes). Results land in pair_consensus table on sma_platform; consensus_status is computed by the existing consensus_builder.py cron.

Compute infrastructure failure prevents any model from completing on any sub-pair by 2026-08-04 (e.g., Spark #2 lost beyond hard-drain, Dell tunnel persistently down, NPLexer Modal weights never downloaded). Treated as null/inconclusive — re-prediction filed Q4.

Third of three Q3-2026 quarterly registrations. Tests whether the platform can produce a CONFIRMED ligand pose against the myostatin x ActRIIB receptor complex - the apitegromab-class axis with an open small-molecule opportunity.

By 2026-08-04 23:59 UTC, the 4-model cross-consensus engine will produce CONFIRMED status on at least one canonical sub-pair: Myostatin prodomain x small-molecule antagonist, ActRIIB ECD x myostatin x small-molecule blocker, or Follistatin x myostatin x stabilizing small molecule.

Each sub-pair queued through Boltz-2 (ligand), AF3 ligand-aware, Chai-1, NPLexer-3. Aggregated by consensus_builder.py.

Compute capacity loss prevents evaluation; re-prediction filed for Q4-2026.

Second of three Q3-2026 quarterly registrations. Tests the platforms RNA-protein-ligand ternary capability moat against the risdiplam-class SMN2 splicing modulator interface.

By 2026-08-04 23:59 UTC, the 4-model cross-consensus engine will produce CONFIRMED status on at least one canonical sub-pair: SMN2 pre-mRNA x U2AF2 x risdiplam-class, SMN2 ISS-N1 x U1A snRNP x small molecule, or SMN2 exon-7 x SRSF1 x splice-modifier.

Each sub-pair queued through Boltz-2 (ligand+RNA), AF3 ternary mode, Chai-1, NPLexer-3. Aggregated by consensus_builder.py.

Compute capacity loss prevents evaluation; re-prediction filed for Q4-2026.