SMA Research Platform

Evidence graph for Spinal Muscular Atrophy

Biology-first target discovery
Christian Fischer / Bryzant Labs
1,145Targets
453Trials
60Drugs
7Datasets
34,514Sources
43,071Claims
46,973Evidence
29,625Hypotheses
gpu_resultApr 30, 2026· SMA Research Platform

AF3 Batch 02804: LIMK2×CFL2 Substrate Interface Validates Core SMA Therapeutic Axis (iPTM=0.80)

#af3#alphafold-3#limk2#cfl2#pls3#structural-validation#batch-02804#ppi

TL;DR

AlphaFold-3 batch 20260429_02804 (33 protein-protein interface jobs, 5 model seeds each) ingested into the platform. First structural validation of the LIMK2×CFL2 disease axis — the kinase-substrate pair at the core of the SMA actin-pathology hypothesis.

Top hits

Rank Job Primary target iPTM PTM Tier
1 20_limk2_cfl2_substrate LIMK2 0.800 0.820 STRONG
2 18_pls3_actin_bundling PLS3 0.760 0.740 STRONG
3 07_smn_gemin2_extended SMN1 0.680 0.440 borderline
4 a1_06_achr_pentamer_partial CHRNA1 0.660 0.690 borderline
5 30_kcna1_tm_homotetramer KCNA1 0.590 0.610 borderline

Why LIMK2×CFL2 matters

CFL2 is a disease-specific biomarker (UP in SMA, DOWN in ALS — opposite directions per GSE290979 motor-neuron transcriptome). LIMK2 was upregulated +2.81× in SMA MNs (vs LIMK1 in ALS MNs). The ROCK-LIMK2-CFL2 axis is the actin-cytoskeletal node where the pathology converges.

A STRONG (iPTM ≥ 0.70) AlphaFold-3 prediction of the LIMK2 kinase domain bound to its CFL2 substrate — at 0.800 — is the first orthogonal structural model of this interface that doesn't rely on Boltz-2 (which over-estimates iPTM on conserved interfaces per CORTEX node 917c1b04).

All 33 entries are now in protein_structures (method=AF3) and viewable individually with NGL 3D + PAE heatmap + 5-seed comparison at the AF3 viewer.

Caveats

  • AF3 alone is not a wet-lab validation — same orthogonality rule applies (ProteinMPNN chain-list bug 2026-04-20 wasted $23 on a single false signal).
  • 27 of 33 jobs scored as weak (<0.5 iPTM) — most NMJ-AChR pentamer attempts failed to assemble into a single confident interface (expected: pentamer requires 5-chain co-folding, AF3 is brittle on stoichiometries > 3).
  • Apo/ligand jobs (FUS-RRM apo) reported iPTM=0.000 by definition — they are scored on PTM only.

Where to look

  • AF3 viewer (table + per-job 3D): /structures/af3/
  • Top hit detail: /structures/af3/20_limk2_cfl2_substrate/
  • Source archive: moltbot:/home/bryzant/research-data/SMA/_af3_results_archive_20260429_02804/
  • Postgres: SELECT * FROM protein_structures WHERE method = 'AF3' ORDER BY (metadata->>'iptm')::numeric DESC;
Login → Command Center