ALS Drug Perturbation — Subtype-Specific Candidate Screening

Repository accompanying the manuscript submission to Brain on blood-derived molecular subtypes of ALS (C1/ALS-Blood-TE, C2/ALS-Blood-Ox, C3/ALS-Blood-Immune, plus a Common pan-ALS signature) and candidate drugs that reverse their gene expression signatures.

• PI: Junguk Hur, University of North Dakota
• Collaborators: Eva Feldman, Masha Savelieff, Yue Zhao (University of Michigan)
• Last updated: 2026-04-24

What's in this repository

This snapshot contains the v5.2 presentation (PPTX + PDF) plus the analysis outputs and figures that accompany the 2026-04-24 update responding to colleague review comments on the v5 deck.

Presentation

File	Description
ALS_Drug_Perturbation_Presentation_v5.2_2026-04-24.pptx	21-slide deck; each slide includes speaker notes citing source data files
ALS_Drug_Perturbation_Presentation_v5.2_2026-04-24.pdf	PDF export of the same deck

Analysis tables (analysis/revised_2026-04-24/)

File	Description
R_enrichment_full.xlsx	GO BP/MF/CC + KEGG + Hallmark + Reactome ORA (clusterProfiler, BH-FDR) for C1/C2/C3/Common × UP/DOWN/ALL DEG signatures
R_enrichment_top.csv	Top-5 pathways per subtype per library
R_enrichment_summary.md	Narrative summary of enrichment results
subtype_hallmark_full.xlsx	Python hypergeometric ORA cross-check against MSigDB Hallmark 2020
subtype_hallmark_enrichment.csv	Top-5 Hallmark hits per subtype (Python cross-check)
subtype_pathway_summary.md	Narrative (Python)
gene_vs_pathway_overlap.csv	Gene-level Jaccard vs pathway-level Jaccard between subtype pairs
gene_vs_pathway_overlap.md	Key result: gene Jaccard 0.46–0.51 vs Reactome top-10 Jaccard = 1.00 across all subtype pairs
subtype_specific_overlap.csv	L1000CDS2-specific × Clue.io-specific drug overlap per subtype
subtype_specific_overlap.md	Narrative
l2_specific_with_clue_context.csv	Per-drug Clue.io NCS context for each L1000CDS2 subtype-specific drug
MSigDB_Hallmark_2020.gmt	Hallmark pathway library used for Python ORA and drug-target Hallmark assignment

Figures

File	Description
figures/networks/clueio_drug_target_{C1,C2,C3,combined}.{png,pdf}	NEW — Clue.io drug–target networks (top-15 reversers per subtype)
figures/networks/drug_gene_network_{C1,C2,C3,combined}.{png,pdf}	L1000CDS2 drug–DEG bipartite networks, regenerated with readable drug labels
figures/heatmaps/drug_connectivity_heatmap.{png,pdf}	Clustered connectivity-score heatmap (39 multi-subtype drugs), readable drug labels
figures/final/drug_convergence_venn.{png,pdf}	L1000CDS2 subtype-Venn (16 C1-only / 23 C2-only / 22 C3-only)

Correspondence (logs/)

File	Description
colleague_reply_2026-04-24.md	Long reply addressing 4 review comments on v5 slides
colleague_reply_short_2026-04-24.md	Email-ready short version

Key findings (2026-04-24 update)

1. Pathway-level convergence explains pan-subtype drug hits. The C1/C2/C3 DEG signatures overlap ~50% at the gene level but the top-10 Reactome pathways are 100% identical across all subtype pairs. Pan-subtype reversers (ritonavir, YM-155, sulfadoxine) act on this shared translation/ribosome/NMD axis.
2. Withaferin-a shows a real subtype gradient (C3 #2, C2 #11, C1 #33 among named compounds), consistent with its IKK/NF-kB mechanism and the UP-skewed inflammatory character of the C3 signature.
3. L1000CDS2 subtype-specificity is largely a top-50 truncation artifact. Only 2 of 60 L1000CDS2 "subtype-specific" drugs preserve their specificity in Clue.io's full 33,609-drug universe (tosedostat in C1, glipizide in C2).
4. All drug labels in figures now include human-readable names in addition to BRD codes (e.g., "lovastatin (BRD-A70155556)").

Data sources

• Clue.io CMap (Broad Institute LINCS L1000)
• L1000CDS2 (Ma'ayan Lab)
• TxGNN knowledge graph (Zitnik Lab)

Citation

Manuscript in preparation. Contact Junguk Hur (jung.hur@und.edu) for citation details.