Integrative multi-phase transcriptomics for precision diagnostics and therapeutic target discovery in amyotrophic lateral sclerosis

Amyotrophic lateral sclerosis (ALS) lacks routine molecular diagnostics, and diagnostic delays averaging twelve to fifteen months from symptom onset substantially erode the therapeutic window. Peripheral blood transcriptomics offers a minimally invasive route toward earlier diagnosis, yet discovery is heavily challenged by cross-platform inconsistency, small peripheral transcriptomic signals, and a lack of biologically grounded validation. Most published signatures are evaluated only against healthy controls, leaving their performance against clinically similar mimic conditions uncharacterized. This study aimed to develop a computational framework for peripheral blood biomarker discovery in ALS and validate the resulting signature across independent cohorts and disease mimics.

We developed a systems-integrated, leakage-free transcriptomic framework using GSE112676 (n = 702) as the sole discovery cohort. Differential expression analysis and weighted gene co-expression network analysis (WGCNA) were performed, followed by a triple-consensus machine learning pipeline combining Random Forest, LASSO regression, and SVM-RFE. A novel Biomarker Priority Score (BPS) integrating cross-cohort diagnostic stability with network centrality was developed for biomarker ranking. The trained model was validated across five independent arms spanning three external cohorts on two microarray platforms and one RNA-seq dataset.

Differential expression analysis identified 976 ALS-associated genes, and WGCNA resolved nine co-expression modules with the brown module showing the strongest disease association (r = 0.368, p < 0.0001). The pipeline yielded a six-gene diagnostic signature (BRI3, ABCA1, QPCT, PPP2R5A, ETFRF1, and SLC37A3) with a discovery area under the curve (AUC) of 0.868 (95% CI: 0.835–0.897) and an independent validation AUC of 0.746 (GSE112680, n = 301). The BPS ranked ABCA1 as the lead candidate. Application of the trained model to ALS mimic patients showed that the signature detects a shared peripheral immune-metabolic response across motor neuron disease conditions rather than an ALS-specific signal (ALS vs mimic: AUC = 0.525; mimic vs control: AUC = 0.717). Furthermore, peripheral immune deconvolution identified two distinct ALS immune subtypes, and drug repurposing analysis nominated clonidine and fingolimod as candidate therapeutic agents.

This study identifies a reproducible peripheral blood transcriptomic signature for ALS and shows that its performance against mimic disorders is reduced, consistent with prior blood transcriptomic studies indicating that mimic discrimination is substantially more difficult than ALS-versus-control classification.