A sample due diligence case from the perspective of PlexoA’s scientific advisory team.
Background
An AI-driven drug discovery startup had generated investor buzz by claiming to identify novel therapeutic targets for neurodegenerative diseases using machine learning on multi-omic datasets. The targets were said to be “biologically novel, mechanistically relevant, and druggable.”
But when one of the early investors began considering a larger follow-on investment, concerns surfaced: were these targets genuinely translational, or simply computationally interesting?
Before advancing, the investor requested a science-led, target-level validation review to determine whether the AI-identified leads had real-world therapeutic potential — or if they were still too speculative for serious development.
The Questions We Asked
When AI meets drug discovery, the promise is great — but biological reality is non-negotiable. Our team at PlexoA focused on five critical questions:
- Is there biological plausibility connecting these targets to disease pathology?
- Do the targets have independent support in peer-reviewed literature or human datasets?
- Are they novel, or just rediscovered low-priority genes?
- How tractable and druggable are these proteins — structurally and pharmacologically?
- Are there translational bottlenecks or red flags that might hinder clinical development?
In the world of biotech, success is not just about groundbreaking science; it’s about translating that science into scalable solutions that can change lives.
What We Evaluated
✅ Biological Plausibility
We reviewed each of the startup’s top 10 AI-nominated targets (across Alzheimer’s and Parkinson’s) for mechanistic relevance, tissue expression, pathway alignment, and knockout phenotypes.
- Several targets showed strong theoretical links to neuroinflammation and protein aggregation pathways.
- Others were poorly characterized, with limited known function in CNS tissue.
✅ Literature Support
We conducted a PubMed, OpenTargets, and GWAS catalog scan for each target.
- Some targets had extensive backing from animal models and pathway studies.
- Others had minimal mentions or functional characterization, signaling high risk.
✅ Innovation Assessment
We assessed whether these targets were truly novel or just echoing existing efforts.
- Approximately 40% of the targets were absent from current clinical pipelines.
- However, a few appeared in competitors’ early IP filings, raising potential freedom-to-operate concerns.
✅ Druggability Review
We reviewed domain structures, known ligands, and previous attempts to modulate the target class.
- Orphan proteins and intracellular scaffolds lacked defined binding pockets or assay systems.
- Kinases and GPCRs scored well.
Key Findings
Interesting Targets — But Uneven Validation
The startup’s AI platform surfaced some promising, underexplored targets. However, the depth of evidence and druggability varied widely. A few were clearly viable for follow-up; others were preliminary hypotheses in need of wet-lab proof.
Red Flags We Flagged
- Biological black boxes: Some high-ranked targets had unknown or unclear physiological roles.
- Druggability gaps: Several targets lacked structural data, ligandability studies, or suitable assays.
- Overstated novelty: One target had an active patent application filed by a large pharma player.
- Translational risk: In vitro-only validation for most targets, with no human tissue confirmation.
What Happened Next
The investor put the brakes on a major follow-on round. Instead, they gave the startup a clear, milestone-driven checklist:
- Validate top 2–3 targets with functional assays and CRISPR knockouts
- Prioritize targets with known tractability or pharmacological precedent
- Begin developing proof-of-concept screens for lead modulation
This preserved the investor’s enthusiasm — while grounding expectations in scientific feasibility, not algorithmic promise.
Leave a Reply