SMA Research Platform

Research directions under active exploration — from spatial multi-omics to engineered probiotics. Each direction connects to specific molecular targets and therapeutic modalities.

Genes, proteins, and pathways implicated in SMA pathogenesis, scored across multiple evidence dimensions. The platform tracks 58 molecular targets across three tiers: Primary targets — SMN1 and SMN2, the causal genes of SMA where loss of full-length SMN protein drives motor neuron degeneration. Established modifier targets — STMN2 (axonal maintenance), PLS3 and NCALD (natural protective modifiers identified in asymptomatic SMN1-deletion carriers), UBA1 (ubiquitin pathway), and CORO1C (actin dynamics). Discovery targets — recently identified through multi-omics convergence analysis and cross-disease research, including PFN1 (profilin, SMA-ALS convergence), CFL2 (cofilin, actin rod formation), ROCK2 (Rho-kinase, druggable with fasudil), and TP53 (p53-mediated motor neuron death). Each target is scored on evidence depth, source diversity, druggability, and clinical validation.

SMA clinical trials aggregated from ClinicalTrials.gov via the v2 API with automated daily refresh. Covers all interventional and observational studies related to spinal muscular atrophy — from early Phase 1 safety trials through Phase 3 efficacy studies and post-marketing surveillance. Each trial entry includes NCT identifier, phase, enrollment, status, intervention type, primary/secondary outcome measures, and where available, published results with adverse events and participant flow data. Use the filters to explore by phase, status, intervention type, or keyword.

Approved SMA therapies and pipeline candidates tracked with mechanism of action, clinical status, and computational screening data. Three FDA/EMA-approved treatments target the SMN pathway directly: nusinersen (antisense oligonucleotide, intrathecal), risdiplam (small molecule splicing modifier, oral), and onasemnogene abeparvovec (AAV9 gene therapy, one-time IV). Pipeline drugs explore complementary approaches including muscle-enhancing (apitegromab, anti-myostatin), neuroprotective (fasudil, ROCK inhibition), HDAC-mediated SMN2 upregulation (panobinostat, vorinostat), and pathway-corrective strategies targeting actin dynamics, p53-mediated apoptosis, and axonal transport. Each drug entry links to DiffDock virtual screening results where available.

PubMed papers and patent literature ingested via automated daily pipeline, each scanned by AI (Gemini Flash with Groq fallback) for structured claims about SMA biology, molecular targets, and therapeutic approaches. The ingestion pipeline runs daily at 03:00 UTC, querying PubMed, bioRxiv/medRxiv preprints, ClinicalTrials.gov, and Google Patents. Each abstract passes a two-layer quality filter: first an SMA-relevance gate (must mention SMA, SMN, motor neuron, or approved therapy names), then a post-extraction quality gate that rejects claims about unrelated diseases. Sources are linked to their extracted claims — use the "With claims" filter to see which papers have been processed.

Curated omics datasets for SMA research. Tier 1 datasets are directly usable for motor neuron vulnerability analysis; Tier 2-3 require additional QC or serve as validation.

Structured scientific assertions extracted from paper abstracts using multi-LLM analysis with rigorous quality filtering. Each claim is a single factual statement that preserves the original authors' hedging language (e.g., "may regulate" stays "may regulate" — never upgraded to definitive). Claims are typed into 12 categories (gene expression, protein interaction, drug efficacy, splicing event, biomarker, etc.), scored for confidence (0–100%), and linked to both their source paper and relevant molecular targets via 200+ alias patterns. The extraction pipeline uses a two-layer quality gate: disease-relevance filtering removes non-SMA contamination, and word-boundary matching prevents false target links. Click any row to see the full provenance chain: paper title, PubMed ID, abstract excerpt, extraction model, and metadata.

Phase 2: Multi-criteria ranked hypotheses scored across evidence depth, source convergence, therapeutic clarity, target strength, and novelty. Tier A = top 5 high-conviction, Tier B = medium priority, Tier C = needs more evidence.

Evidence-grounded, falsifiable predictions generated from convergence scoring across 5 dimensions: Volume, Lab Independence, Method Diversity, Temporal Trend, and Replication. All scoring weights are transparent methodology. Each card links every claim to its source paper.

The platform's highest-confidence mechanistic finding: SMN protein loss triggers a cascade through the actin dynamics axis — PFN1 dysregulation activates ROCK2, which through LIMK1 locks Cofilin (CFL2) in its inactive phosphorylated state, causing toxic actin-cofilin rods that block axonal transport and ultimately destroy the neuromuscular junction. A parallel apoptotic branch via p38/MAPK14 activates p53-mediated motor neuron death. Both axes are druggable with existing compounds. Click any node for evidence details and live claim counts.

Multi-dimensional evidence convergence scoring across thousands of curated, quality-filtered claims extracted from 6,400+ PubMed sources. Each of the 58 molecular targets is scored across five independent dimensions: Claim Volume (raw evidence mass — how many distinct assertions support this target), Lab Independence (number of unique research groups reporting findings — guards against single-lab bias), Method Diversity (range of experimental approaches: in vitro, animal model, patient data, computational — cross-validated findings score higher), Temporal Trend (whether evidence is growing, stable, or declining over recent years — captures scientific momentum), and Replication (how often key findings have been independently confirmed across different studies and model systems). Scores are weighted and combined into a composite convergence score (0–100). All weights and methodology are fully transparent and transparent methodology. The engine generates falsifiable predictions grounded in evidence — each prediction card links every supporting claim back to its source paper for full traceability.

Bayesian back-testing of convergence scores against known drug outcomes — the critical self-check that separates rigorous research from speculation. For each drug with a known clinical outcome (approved, failed in Phase 2/3, or preclinical only), the platform asks: did our evidence scoring predict the right outcome? The calibration process works as follows: (1) Outcome collection — gather real-world drug approval/failure data from ClinicalTrials.gov and FDA records for all 21 tracked drugs. (2) Score comparison — compare each drug's convergence score against its actual clinical outcome. Approved drugs (nusinersen, risdiplam, onasemnogene) should score high; failed drugs should score low. (3) Bayesian updating — use the comparison to compute posterior probabilities, measuring how well evidence mass predicts clinical success. (4) Grade assignment — the system earns a calibration grade (A–F) based on concordance between predicted and actual outcomes. Grade A (current: 89.8%) means the scoring reliably separates successful from unsuccessful therapeutic approaches. A well-calibrated platform means researchers can trust the convergence scores when evaluating novel, untested targets.

Multi-criteria scoring across 7 dimensions: evidence strength, biological coherence, fragility relevance, interventionability, translational feasibility, novelty, and contradiction risk. Composite score determines Phase 3 priority.

Multi-criteria decision engine integrating 6 data dimensions: evidence convergence (25%), druggability via DiffDock screening (20%), ESM-2 structural uniqueness (15%), clinical validation from drug outcomes (15%), cross-species conservation (10%), and target novelty (15%).

The evidence graph connects claims to their supporting sources. Each assertion is backed by traceable references (PMIDs, clinical trial results). Grouped by source paper, sorted by claim count.

This pipeline computationally filters thousands of ChEMBL compounds down to the best candidates for SMA drug discovery. The process runs in six steps: (1) ChEMBL query — compounds bioactive against top-scored SMA targets are fetched with their SMILES strings; (2) RDKit descriptor calculation — molecular weight, LogP, rotatable bonds, H-bond donors/acceptors, TPSA, and QED are computed from SMILES; (3) Lipinski Rule of 5 — MW < 500, LogP < 5, HBD ≤ 5, HBA ≤ 10; compounds failing two or more rules are flagged as non-drug-like; (4) BBB permeability estimate — TPSA < 90 Ų and MW < 450 are used as a heuristic for blood-brain barrier crossing; (5) CNS MPO score — a 0–6 composite of LogP, LogD, MW, TPSA, HBD, and pKa tuned for CNS drug development; (6) PAINS filter — substructure alerts for pan-assay interference compounds that cause false positives in biochemical screens.

Why BBB penetration matters for SMA: SMA is caused by the loss of SMN protein in lower motor neurons located in the anterior horn of the spinal cord — a compartment behind the blood-brain barrier. Small-molecule therapeutics must cross this barrier to reach motor neurons. Risdiplam (approved 2020) succeeds partly because of its BBB-permeable profile; many otherwise potent compounds fail in SMA because they cannot access the CNS. Compounds with TPSA > 90 Ų or MW > 500 Da are unlikely to achieve meaningful CNS exposure via oral dosing.

Score glossary: Lipinski — binary pass/fail for oral bioavailability potential. BBB — heuristic estimate of CNS penetration (TPSA + MW). CNS MPO — 0–6 score; ≥ 4 is considered CNS-optimized. QED — 0–1 drug-likeness estimate combining eight Lipinski-adjacent properties; ≥ 0.5 is high quality. PAINS — substructure alert for reactive or promiscuous scaffolds that should be deprioritized.

Note: Drug-likeness predictions use rule-based heuristics (Lipinski Rule of 5, TPSA-based BBB estimate, QED score). These are filtering tools, not validated PK/tox models.

Top Candidates

Drug repurposing means finding new therapeutic uses for existing approved drugs — bypassing the 10–15 years and $1–2B typically required for de novo drug development. Repurposed drugs have already passed safety trials, so clinical translation is dramatically faster: Phase I is often skipped and Phase II can start in 2–3 years rather than 10+.

The platform identifies SMA repurposing candidates through three convergent strategies: (1) Cross-disease mining — drugs approved or in trials for related neuromuscular diseases (ALS, Duchenne Muscular Dystrophy, SBMA, CMT) that share molecular targets with SMA; (2) ChEMBL bioactivity — known compounds with high pChEMBL values (≥ 6.0, corresponding to IC₅₀ ≤ 1 µM) against top-scored SMA targets; (3) Pathway overlap — compounds whose known mechanism overlaps with the actin dynamics, NMJ signaling, or autophagy/survival pathways dysregulated in SMA.

Precedent in SMA: Valproic acid (VPA), originally an epilepsy drug, was one of the first compounds tested in SMA clinical trials — its HDAC inhibition was found to increase SMN2 splicing. Olesoxime (a cholesterol-oxime neuroprotective) reached Phase II. Riluzole (ALS-approved) showed modest motor neuron protection in SMA models. The platform extends this approach computationally, scoring each candidate 0–1 based on target relevance, potency, clinical phase, and pathway convergence. Click any row to see full rationale, mechanism, and target link.

Top Candidates

In drug discovery, a hit is a compound that shows measurable activity against a target of interest and passes initial computational filters. This section is the unified ranked list — the best compounds from all analysis pipelines: ChEMBL screening, cross-disease repurposing, and DiffDock virtual binding.

Each candidate passes through a 6-stage validation pipeline: (1) Computational — drug-likeness filters (Lipinski, QED, PAINS), BBB/CNS MPO scoring; (2) Structural — DiffDock pose prediction against SMA target binding pockets, confidence scoring; (3) Analog search — ChEMBL SAR analysis to identify structurally similar compounds with known SMA-relevant activity; (4) ADMET prediction — rule-based absorption, distribution, metabolism, excretion, and toxicity estimates; (5) Literature review — automated PubMed search for the compound + SMA target co-occurrence; (6) Experimental design — suggested assay types (SMN2 splicing reporter, NMJ morphology, motor neuron survival) for wet-lab validation.

Candidates are scored 0–1 (integrated score) and assigned a tier: Tier A (≥ 0.6) — strong multi-dimensional evidence, prioritized for experimental follow-up; Tier B (0.4–0.6) — moderate evidence, worth secondary screening; Tier C (< 0.4) — computational-only signal, lower priority. Click any row to see full molecular properties, BBB status, DiffDock score, validation stage, and target link.

Note: ADMET predictions use rule-based heuristics (Lipinski Rule of 5, TPSA-based BBB estimate, QED score, PAINS substructure filters). These are computational filtering tools, not validated pharmacokinetic or toxicology models.

Ranked Candidates

Positive binding predictions from AI-driven virtual screening. Each hit goes through a 6-stage validation pipeline: computational validation, structural analysis, analog search, ADMET prediction, literature review, and experimental design.

Note: ADMET predictions in the pipeline use rule-based heuristics (Lipinski, TPSA, PAINS), not validated PK/tox models.

Interactive network of SMA molecular targets connected by protein-protein interactions (STRING), shared pathways (KEGG/UniProt), and compound bioactivity (ChEMBL). Click a node to highlight its connections.

Structured database of drug successes and failures in SMA research. Every outcome traces back to a source paper — capturing not just what worked, but why compounds failed (toxicity, bioavailability, efficacy).

Cross-species conservation mapping of SMA-relevant molecular targets across 7 model organisms. Each organism offers unique advantages for SMA research: mice (Mus musculus) serve as the primary disease model with SMN-delta7 and Taiwanese SMA strains; zebrafish (Danio rerio) enable rapid drug screening with motor neuron fluorescent reporters; the naked mole rat (Heterocephalus glaber) shows exceptional neuronal resilience and resistance to neurodegeneration; axolotl offers complete spinal cord regeneration. Conservation scores are computed from NCBI Ortholog data — a score of 71% or higher indicates strong evolutionary conservation, suggesting the target's function is preserved across species and findings from model organisms are likely translatable to humans. Click any species card to see which SMA targets have orthologs in that organism, or click any heatmap cell to view the specific ortholog with links to NCBI Gene and STRING-DB.

Conservation Heatmap (click cells for details)

16 research directions spanning spatial multi-omics, regenerative biology, and computational approaches to SMA. Click any direction to see connected targets, claims, and hypotheses.

Essential databases, tools, registries, and organizations for SMA researchers. Open as standalone page

Ask any question about SMA research or search across thousands of evidence claims and thousands of sources. Ask follow-up questions to go deeper.

Generate publication-ready evidence summaries for any SMA target or topic. Powered by Claude Sonnet synthesizing across all platform data (claims, hypotheses, trials, drug outcomes).

Browse 800+ AI-generated and computationally screened molecules for SMA drug targets. Includes MolMIM scaffold decorations, GenMol analogs, DiffDock docking results, and ML-proxy 100k virtual screen hits. Filter by target, drug-likeness, BBB permeability, and more. Export as CSV (researchers) or SDF (chemists).

CRISPR/CRISPRi guide RNA design for SMN2 exon 7 region. Three therapeutic strategies: CRISPRi at ISS-N1 (mimic nusinersen), CRISPRi at ESS (block hnRNP A1), CRISPRa at ESE (enhance Tra2-beta). 20 nt protospacer + NGG PAM, GC 40-70%, polyT filtered.

SMN2 Regulatory Motifs

Top Guides (All Strategies)

AAV serotype evaluation for SMA gene therapy delivery. 9 capsids scored across motor neuron tropism, BBB crossing, immunogenicity (NAb seroprevalence), manufacturing feasibility, and packaging capacity. Zolgensma uses AAV9 (scAAV9-SMN1).

Capsid Rankings

"GitHub for Life" — every SMN2 sequence variant is a deterministic commit with a SHA-256 hash. The disease (SMA) is a single-nucleotide bug (C→T at position 6). Therapeutic edits are patches that restore function. Track the lineage from SMN1 (healthy) through SMN2 (disease) to corrected variants.

Version Tree

Click any row to expand the full sequence diff, clinical significance, and population frequency.

Sequence Diffs

Each diff shows the exact nucleotide changes between parent and child variant. Position numbers refer to the SMN exon 7 coordinate system.

Pharmacophore-based docking score prediction for SMA drug candidates against 7 target binding pockets. Scores compounds from the molecule_screenings database by shape complementarity, H-bond potential, hydrophobic match, electrostatic alignment, and strain penalty.

Top Predicted Binders

Machine learning surrogate trained on 4,116 DiffDock v2.2 results. Uses RDKit Morgan fingerprints (ECFP4, 2048-bit) + RandomForest to predict binding confidence ~1000x faster than physics-based docking. Enables screening millions of molecules on CPU in minutes.

Actual vs Predicted (Training Set)

Top 20 Feature Importances

Target Distribution

Prime editing (PE2/PE3/PEmax) assessment for SMA: SMN2 C6T correction (the root cause fix), ISS-N1 disruption (permanent nusinersen), and ESE strengthening. Compared with approved therapies. Prime editing = reverse transcriptase + Cas9 nickase + pegRNA — no double-strand breaks.

Therapy Comparison

Agent C — Molecular Dynamics simulation code generator. Generates ready-to-run OpenMM Python scripts for SMA-relevant protein simulations: SMN oligomerization, hnRNP A1-ISS-N1 binding, risdiplam mechanism, NCALD calcium dynamics, PLS3 actin bundling, SMN-Gemin2 stability.

Phase 7.1 — Drug penetration modeling across spinal cord microanatomy. Maps which SMA drugs reach which tissue compartments based on molecular properties, BBB permeability, and CSF exposure. Identifies therapeutic "silent zones" where current drugs underperform.

Spinal Cord Zones

Drug Penetration

Silent Zones

Phase 7.2 — Cross-species regeneration programs in axolotl and zebrafish compared with degeneration in human SMA motor neurons. Identifies conserved repair pathways that are silenced in SMA and could be therapeutically reactivated.

Regeneration Genes

Pathway Comparisons

Phase 7.3 — Muscle-to-nerve retrograde signaling at the neuromuscular junction. Tests the "happy muscle → surviving neuron" hypothesis: can improving muscle health rescue motor neurons via retrograde trophic signals?

Retrograde Signals

EV Therapeutic Cargo

Organ-on-Chip Models

Phase 7.4 — SMA is not just a motor neuron disease. Liver, cardiac, metabolic, pancreatic, vascular, skeletal, and GI pathology emerges especially in severe SMA types. Models the full systemic picture and combination therapy strategies.

Affected Organ Systems

Combination Therapies

Each strategy targets multiple disease axes simultaneously. Click a card to see drugs involved, mechanism rationale, and clinical evidence.

Energy Budget Model

SMA motor neurons run an energy deficit: SMN loss impairs mitochondrial function, actin dynamics require ATP, and retrograde transport of neurotrophic factors stalls. The energy budget model compares ATP supply vs. demand across normal, SMA, and treated motor neurons. A supply/demand ratio below 0.7 predicts neurodegeneration.

Phase 7.5 — Ion channel expression, membrane potential (Vmem) states, and electroceutical interventions for SMA motor neurons. Based on Michael Levin's bioelectricity framework: many SMA MNs are alive but electrically dormant — they can potentially be reactivated.

Ion Channels

Vmem States

Electroceuticals

Phase 9.3 — Axolotl and zebrafish use alternative splicing as a master switch for regeneration. The same genes exist in humans but their regeneration-promoting isoforms are epigenetically silenced. This module maps 10 regeneration-specific splice events to human orthologs.

Phase 9.4 — Predicts RNA-binding affinity of compounds toward SMN2 pre-mRNA regulatory elements. 6 RNA target sites mapped (ISS-N1, 5'ss/U1 interface, ESE2, ESS, branch point, TSL2). Benchmarks against known modulators like risdiplam and branaplam.

RNA Target Sites in SMN2

Known SMN2 Modulators

Phase 6.1 — Compounds that simultaneously modify SMN2 splicing AND influence ion channels. The bioelectricity intersection: fixing the gene is not enough — reactivating the electrical function of rescued motor neurons is the missing therapeutic layer.

Phase 10.3 — Multi-scale computational model of the SMA motor neuron. Simulates drug combinations across 5 compartments (soma, axon, NMJ, dendrites, nucleus) and 8 signaling pathways. Predicts synergistic drug combinations in silico.

Motor Neuron Compartments

Signaling Pathways

Optimal Drug Combinations

Phase 10.4 — Open-source experiment design automation. 8 standardized SMA assays with timeline and protocol specifications. 3 cloud lab integrations (Emerald Cloud Lab, Strateos, Opentrons). Generates complete experiment designs from hypothesis text.

SMA Assay Library

Cloud Lab Integrations

Phase 10.5 — Zero-knowledge data sharing framework for SMA research. Enables cross-institutional collaboration without sharing raw patient data. Federated learning protocols, OMOP/OHDSI data model mapping, privacy budget calculator, and 4-tier data sharing framework.

Federated Learning Protocols

Data Sharing Tiers

OMOP/OHDSI Mappings

Phase 11 — Translating platform discoveries into real-world impact. Regulatory pathway mapping (FDA/EMA), grant application templates, and a 5-level hypothesis validation pipeline from computational validation to IND filing.

Regulatory Pathways

Grant Templates

Validation Pipeline

Gold-standard computational predictions from DiffDock, SpliceAI, ESM-2, and Cas-OFFinder. Every result is traceable to its tool version, parameters, and input data. View GPU scripts on GitHub →

Questions about the platform, data, or collaboration? Send us a message.

Research highlights, computational discoveries, and platform updates. Each post documents a specific finding with full methodology and source citations. Comment on findings and join the discussion.

AlphaFold2 and ESMfold predicted 3D structures for all SMA research targets. Each structure is scored by pLDDT (predicted Local Distance Difference Test) — a per-residue confidence metric: green (≥85) = high confidence, well-ordered; amber (≥70) = moderate confidence; gray (<70) = low confidence, likely disordered. High-confidence structures are most suitable for structure-based drug design and docking. Click any row for AlphaFold DB / UniProt links and linked discovery data.

Why this matters for SMA drug design: Three-dimensional protein structure directly determines which pockets small molecules can bind. Before computational structure prediction, only ~10% of proteins had experimentally solved structures. AlphaFold v6 and ESMfold now provide high-accuracy models for nearly all human proteins, enabling virtual screening and binder design even for targets without crystallographic data.

Structures predicted via AlphaFold DB v6 (EMBL-EBI), ESMfold v1, and Boltz-2 (Chai Discovery). pLDDT scores from predicted structures. Pre-existing PDB structures retain original experimental resolution.

P2Rank-predicted binding pockets across 58 SMA target protein structures. Each protein is scored by pocket druggability probability (0-1) and pocket score. Highly druggable (probability ≥ 0.8) targets have well-defined binding sites suitable for small molecule drug design. Druggable (0.5-0.8) targets may require optimized ligands. Marginal (< 0.5) pockets are shallow or solvent-exposed. Click any row to expand individual pocket details including residue composition and 3D center coordinates.

Pocket predictions via P2Rank 2.5.1 with AlphaFold-optimized configuration. Druggable flag requires score > 50 AND probability > 0.8.

ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) predictions for 21,000+ compounds across all SMA targets. Compounds are scored for drug-likeness (QED), blood-brain barrier permeability (BBB), CNS Multi-Parameter Optimization (MPO), Lipinski Rule-of-Five compliance, and physicochemical properties (MW, LogP, TPSA, HBD, HBA). Filter by properties to identify the most promising CNS-penetrant drug candidates for SMA.

Non-obvious connections across thousands of curated claims from different papers — the platform's core differentiator. While individual papers report isolated findings, cross-paper synthesis reveals hidden patterns: targets that co-occur in unrelated studies, shared mechanisms between seemingly independent pathways, and transitive bridges (if Paper A links X→Y and Paper B links Y→Z, the platform discovers X→Z). The analysis builds a co-occurrence matrix across all claims, identifies statistically significant target pairs, and generates synthesis cards that explain the biological connection with full citation trails. This is how the platform discovered the ROCK-cofilin-actin rod pathway as a therapeutic axis — no single paper described the complete pathway, but the synthesis engine connected findings from 12+ independent publications.

AI-predicted drug-target synergy scores combining docking affinity, literature evidence, pathway overlap, and claim support. Identifies the most promising multi-mechanism therapeutic combinations for SMA.

DiffDock v2.2 docking predictions. Extended campaign: 224 dockings across 8 targets (ROCK2, MAPK14, LIMK1, SARM1, and more), plus 378-compound batch screen. View protein binders and AI-generated molecules in GPU Results →

Auto-generated comprehensive research summary for external collaborators, professors, and grant reviewers.

Real-time summary of platform capabilities, evidence depth, and research progress.

What this platform has computed since launch. Live numbers from the database, factual milestones, and infrastructure used.

Today's Numbers

Growth Timeline

Pipeline Stats

Computational Resources Used

Query our evidence graph programmatically. No authentication required for read access. All endpoints return JSON under /api/v2.

Core Data Endpoints

Computational Biology

Data Export

Claim Type Reference

Python Example

import requests

BASE = "https://sma-research.info/api/v2"

# Get scored and ranked targets
scores = requests.get(f"{BASE}/scores", params={"mode": "discovery"}).json()

for t in scores[:10]:
    print(f"{t['symbol']:10s} score={t['composite_score']:.3f}")

# Search high-confidence drug efficacy claims
claims = requests.get(f"{BASE}/claims", params={
    "claim_type": "drug_efficacy",
    "confidence_min": 0.8,
    "enriched": True,
    "limit": 100
}).json()

for c in claims:
    print(f"[{c['confidence']:.2f}] {c['predicate'][:80]}")

# Deep-dive: full evidence for a target
target = requests.get(f"{BASE}/targets/symbol/ROCK2").json()
dive = requests.get(f"{BASE}/targets/{target['id']}/deep-dive").json()
print(f"Claims: {len(dive['claims'])}, Hypotheses: {len(dive['hypotheses'])}")

R Example

library(httr)
library(jsonlite)

base_url <- "https://sma-research.info/api/v2"

# All targets with discovery-mode scores
scores <- fromJSON(content(
  GET(paste0(base_url, "/scores"), query = list(mode = "discovery")),
  "text"
))

# Top 10 by composite score
top10 <- head(scores[order(-scores$composite_score), ], 10)
print(top10[, c("symbol", "composite_score")])

# Export as CSV
resp <- GET(paste0(base_url, "/export/targets"), query = list(fmt = "csv", limit = 5000))
writeLines(content(resp, "text"), "sma_targets.csv")

Rate Limits & Access

Citation

Fischer, C. (2026). SMA Research Platform — Open Evidence Graph
for Spinal Muscular Atrophy. https://sma-research.info
Bryzant Labs. Accessed [date].

@misc{fischer2026sma,
  author = {Fischer, Christian},
  title = {{SMA Research Platform --- Open Evidence Graph for SMA}},
  year = {2026},
  url = {https://sma-research.info},
  note = {Accessed: 2026-03-25}
}

Try It Live

Select an endpoint and click Send to try the API.

Full documentation: Swagger UI | ReDoc | Last updated: 2026-03-25