Technical Due Diligence

Biotech due diligence requires a fundamentally different skill set than financial or commercial assessment. A discovery platform can look impressive in a pitch deck—proprietary libraries, AI-driven screening, novel scaffolds—while having significant technical limitations that only become apparent when you understand the underlying science. The job of technical due diligence is to separate genuine capability from marketing.

For antibody or nanobody discovery platforms, the key questions center on library quality and diversity. What is the actual functional diversity of the library—not the theoretical combinatorial space, but the number of unique, correctly folded clones? What hit rates does the platform deliver against validated targets, and how do those hits perform in affinity characterization? A platform that delivers 50 unique binders per campaign with K_D values in the low-nanomolar range is genuinely useful. A platform that reports 500 "hits" by ELISA but only 5 survive SPR confirmation has a specificity problem.

For expression and manufacturing platforms, ask about scale-up history. Has the system produced material at the gram scale? Has it been used to manufacture material for in vivo studies or clinical trials? Many early-stage platforms work at the microgram scale but encounter serious problems—aggregation, host cell protein contamination, glycosylation heterogeneity—when scaled to manufacturing-relevant volumes.

AI-driven drug discovery has attracted enormous investment, and with it, enormous hype. Evaluating computational claims requires asking specific, technical questions. What training data was used, and is there a risk of data leakage between training and test sets? What are the model's performance metrics on held-out data that the company did not use during development? Has the model made prospective predictions that were validated experimentally, or are all reported results retrospective?

The strongest evidence for a computational platform is a track record of prospective experimental validation: the model predicted X, the team synthesized and tested X, and the experimental result confirmed the prediction. Retrospective benchmarks against public datasets are a necessary starting point but are insufficient on their own—they demonstrate that the model can fit known data, not that it can predict unknown outcomes. Be especially cautious of claims that rely entirely on in silico metrics (docking scores, predicted binding free energies, fold confidence) without experimental confirmation.

Green flags include published data in peer-reviewed journals with clearly described methods, disclosed hit rates and failure rates (not just successes), partnerships with pharma companies that have done their own due diligence, and a team with deep domain expertise in both the biology and the technology. A willingness to share raw data under NDA and to discuss limitations openly is perhaps the strongest positive signal.

Red flags include claims of "platform generality" without target-specific validation data, hit rates reported only as ELISA positives without kinetic confirmation, proprietary metrics that cannot be compared to industry standards, overreliance on in silico validation without wet-lab confirmation, and a team that lacks hands-on experience in the biological domain they are targeting. The most concerning red flag is a company that cannot clearly explain why their approach should work—not just that their numbers look good, but what biological or physical principle underlies their platform's advantage.