In 2021, Verge Genomics, a US-based AI drug discovery company, advanced an ALS candidate into Phase 1 trials. The compound, VRG50635, was not found through conventional screening. It surfaced via the company's CONVERGE AI platform, which analyzed genetic and transcriptomic data from human patient tissue. That trial is ongoing, and no AI-discovered drug for motor neurone disease (MND) has yet reached the market.
The slow progress in MND drug development stems from the disease's biological complexity. Most cases are sporadic, with no clear genetic cause. Among the inherited cases, mutations in the C9orf72, SOD1, TARDBP, and FUS genes are implicated. The protein TDP-43 misfolds in nearly all patients, but the mechanisms remain poorly understood. Traditional drug development relies heavily on animal models that do not fully replicate human disease and has produced only a handful of approved therapies. The FDA approved tofersen (Qalsody) for SOD1-ALS in April 2023, but that drug was developed using antisense oligonucleotide technology, not AI.
AI platforms aim to cut the time and cost of finding viable drug targets by processing large biological datasets that would be impractical for humans to analyze manually. The question is whether these computational approaches can overcome the complexity that has stymied conventional investigation for decades.
The AI Platforms and Techniques
Two companies, Verge Genomics and BenevolentAI, represent the leading approaches to applying AI to MND.
Verge Genomics and CONVERGE
Verge Genomics built CONVERGE, a platform that integrates human genomic data, gene expression profiles, and clinical information. The system uses machine learning to identify genes and pathways that are dysregulated in disease and to predict which compounds might reverse those patterns. Unlike many drug discovery efforts that begin with animal models, CONVERGE is trained on human tissue data, which the company argues produces more relevant leads.
BenevolentAI and the Knowledge Graph
BenevolentAI, a UK-based firm, uses a different technique. Its platform constructs a knowledge graph of scientific literature, patents, trial results, and biological databases. Machine learning algorithms then infer relationships between genes, proteins, diseases, and existing drugs that a human researcher might miss. The system can propose drug repurposing opportunities, where an approved drug for one condition is tested against MND, or identify novel leads. BenevolentAI has published work on potential MND targets, though the company has not disclosed a clinical-stage candidate specifically for the disease.
The Academic Collaborator
The Sheffield Institute for Translational Neuroscience (SITraN) is a major academic collaborator in this space. SITraN has worked with computational biologists to analyze patient-derived data, including stem cell models, to validate leads flagged by AI systems. The institute's role is to provide biological validation for computationally generated hypotheses.
How AI Addresses MND Biology
Untangling Genetic Heterogeneity
Mutations in C9orf72, SOD1, TARDBP, and FUS account for only a fraction of familial cases. The remaining cases involve multiple genes with small effects, which traditional statistical genetics struggles to disentangle. Machine learning models can weigh thousands of genetic variants simultaneously and identify combinations that correlate with disease risk.
Decoding TDP-43 Proteinopathy
TDP-43 misfolds and aggregates in the motor neurons of nearly all MND patients, yet the triggers and downstream consequences are not fully understood. AI platforms can analyze proteomic data to find which cellular pathways are disrupted when TDP-43 aggregates, and then screen for compounds that restore normal function. Verge Genomics has stated that its platform pinpointed leads involved in autophagy and neuroinflammation, processes linked to TDP-43 pathology.
Bypassing Failed Animal Models
Many compounds that worked in mice or rats failed in human trials. By training on human data, AI systems attempt to bypass this gap. Whether human-data-driven models will yield higher success rates in the clinic is not yet proven, but the approach is a deliberate departure from decades of animal-first investigation.
Clinical Milestones and Funding
The VRG50635 Trial
The most concrete validation of AI in MND research is Verge Genomics' VRG50635, which entered Phase 1 trials in 2021. Phase 1 tests safety and dosing in a small number of healthy volunteers and patients. The company has not disclosed whether the trial has advanced to Phase 2 or been halted. The compound is a PIKfyve inhibitor, a lead that CONVERGE flagged as relevant to ALS pathology.
Venture and Public Funding
Verge Genomics raised significant venture funding, including a USD 98 million Series B round in 2021 led by SoftBank Vision Fund 2, though the company does not break out how much of that is earmarked for ALS work specifically. BenevolentAI, which went public via a SPAC merger in 2022, has not disclosed an MND-specific funding figure. Its platform is applied across multiple diseases, and MND is one of several neurology programs.
Academic and Philanthropic Sources
Academic funding for AI-driven MND work has come from sources such as the UK Motor Neurone Disease Association and the National Institutes of Health in the US, but precise totals for AI-specific grants are not publicly aggregated. The Ice Bucket Challenge in 2014 raised global awareness and funding for ALS research, but that money was directed broadly, not specifically to AI approaches.
Limitations and Failures
No Approved AI-Discovered Drug
AI-driven drug discovery for MND has not yet produced a drug that has completed Phase 3 trials or received regulatory approval. The field is still in the validation stage. VRG50635, the most advanced AI-identified candidate, remains in early testing with no published efficacy data.
Data Scarcity and Quality
A fundamental limitation is data quality. AI models are only as good as the datasets they are trained on. Human tissue samples from MND patients are scarce, especially from early-stage disease. Most available tissue comes from post-mortem donations, which reflect end-stage pathology. Models trained on such data may identify leads that are consequences of the disease rather than causes. Longitudinal data from living patients, collected via biomarkers or imaging, is limited.
Reproducibility and Industry Failures
Another limitation is the reproducibility crisis in preclinical work. AI-generated hypotheses must be tested in the lab, and many will fail. BenevolentAI and Verge Genomics have published target discoveries, but independent replication is rare. The pharmaceutical industry has seen high-profile AI drug discovery failures in other therapeutic areas, and MND is unlikely to be an exception. As of May 2024, no AI-discovered MND drug has been approved, and the field remains a high-risk, high-reward bet.
