The Manufacturing Gap Threatening AI Drug Discovery's Promise

The artificial intelligence revolution in drug discovery has encountered an unexpected obstacle: translating digital designs into physical medicines. While algorithms can now design novel proteins and therapeutic candidates in hours, converting those computational blueprints into manufactured drugs requires experimental infrastructure that many AI-first biotechs lack, according to a February 2025 analysis from Amadeus Capital Partners.

The venture firm's portfolio companies illustrate both the promise and practical constraints of computational biology. Healx recently dosed the first patient in a Phase 1/2 trial of HLX-4310 for relapsed and refractory sarcoma, with particular focus on osteosarcoma—a tangible validation milestone for AI-discovered therapeutics. Yet reaching this point required far more than sophisticated algorithms.

From Silicon to Cells: The Validation Bottleneck

The shift from computational design to physical validation represents a fundamental transition in the AI drug discovery sector. Researchers at the Innovative Genomics Institute recently demonstrated this potential by using AI to design gene-editing enzymes that don't exist in nature, with some performing as well as or better than natural enzymes in bacteria, plants, and human cells.

However, generating promising computational predictions is only the beginning. Nuclera has published work addressing how to accelerate the design-build-test-learn cycle for AI-designed antibodies, positioning itself as critical infrastructure for moving AI-generated proteins from digital design to experimental validation. This experimental testing phase generates proprietary data that becomes training material for subsequent AI predictions, creating a competitive advantage for companies with integrated platforms.

Manufacturing: The Ultimate Competitive Advantage

Beyond validation lies an even more substantial challenge: manufacturing at scale. Constructive Bio exemplifies this frontier, developing both novel protein chemistries not found in nature and the manufacturing systems required to produce them commercially. For AI-designed biologics that diverge significantly from natural proteins, existing production infrastructure may prove inadequate.

This manufacturing bottleneck has significant implications for healthcare providers and investors. While AI-discovered therapeutics are beginning to reach clinical trials, the path from algorithm to approved therapy remains lengthy and capital-intensive. Companies focused solely on computational drug discovery without partnerships or internal capabilities for experimental validation and manufacturing scale-up may face competitive limitations.

Broader Diagnostic Applications

The integration of AI with physical infrastructure extends beyond drug discovery. Quibim launched QP-Breast, a CE and UKCA-marked AI tool for radiologists to detect suspected breast lesions on MRI in Europe and the UK. Respiro Diagnostics raised £1 million with Amadeus participation to advance a non-invasive liquid-biopsy platform analyzing DNA, RNA, and proteins from exhaled breath for lung diagnostics.

Victorian Clinical Genetics Services received ISO 15189 accreditation for clinical whole-genome sequencing workflow using Oxford Nanopore technology, demonstrating how AI-enabled diagnostics must still meet rigorous regulatory and quality standards.

The lesson emerging from Amadeus Capital's portfolio is clear: in AI-driven biotechnology, computational innovation must be matched with experimental rigor and manufacturing capability. The companies that successfully bridge the digital-physical divide will likely define the next generation of therapeutic development.

Sources

1. Capitalise #22 - AI has left the chat — Amadeus Capital