Ep174: Najat Khan on the TechBio Movement

Episode Summary

Executive Summary: Najat Khan argues that AI can improve biotech R&D only through disciplined, bilingual teams, high-quality data, and pragmatic use cases—not hype. She explains how she brought AI into J&J, starting with clinical development for faster impact, and why she joined Recursion to help build an AI-native medicines company that integrates discovery, development, and commercial thinking from the outset.

Main Topics: AI in biotech needs pragmatism, not hype (Priority: 5/5): Khan says AI’s potential is real but overpromised. Success depends on where and how it is applied, with humility about limits and a focus on specific pain points rather than sweeping transformation claims. Personal journey and cross-disciplinary identity (Priority: 4/5): She describes growing up in Bangladesh and the UK, moving to the U.S., and developing an early blend of science, coding, debate, and business interests that shaped her multi-silo career path. Why clinical development was the first AI beachhead at J&J (Priority: 5/5): Khan explains that clinical development offered faster time-to-value, large cost impact, and practical use cases such as recruitment, trial design, and patient selection, making it a strategic place to prove AI value. Resistance to AI inside organizations (Priority: 4/5): She details organizational skepticism, fear of replacement, and incentives that protect existing budgets and influence. She argues that adoption requires resilience, trusted leadership support, and high execution standards. Why Recursion appealed as an AI-native platform (Priority: 5/5): Recursion’s first-principles, AI-built-from-the-ground-up model, strong wet/dry lab loop, multimodal datasets, and chemistry acquisitions made it attractive as a company trying to build a full-stack medicines engine. Portfolio strategy: first-in-class and hard-to-drug programs (Priority: 5/5): Khan describes prioritizing programs based on data, differentiation, and patient value, balancing novel targets with known but difficult biology where AI and chemistry can improve therapeutic index and odds of success. Trust, openness, and ecosystem building (Priority: 4/5): She says credibility comes from publishing high-quality science, selectively open-sourcing tools, participating in pre-competitive consortia, and being transparent about what AI can and cannot do.

Key Arguments: AI should be applied where it matters most, with a deep understanding of industry pain points; broad hype alone does not improve drug R&D. Clinical development is a strong entry point for AI because it consumes about 70% of drug development spend and can show value faster than discovery. Trial recruitment is a major bottleneck: if AI can improve site selection and patient identification, it can materially accelerate development timelines. High-quality, reproducible, well-annotated data is more important than simply hoovering up more data; garbage in still produces garbage out. Adoption requires bilingual teams—people who understand both biology/clinical development and machine learning—and strong cross-functional collaboration. Resistance to AI often stems from fear of replacement and from internal power dynamics, not just technical skepticism. Recursion’s appeal lies in being AI-native from the start, with automation, phenotypic screening, multimodal biology, and chemistry design integrated into one system. The most valuable programs are those that are differentiated for patients and judged early on both scientific and commercial viability. Trust is built through disciplined execution, publishing, transparency, and participation in industry-wide standards and consortia.

Data Points: Drug development success rate: ~10% - Khan and the host cite the low overall probability of a clinical compound becoming an actual medicine. Time to make a drug: 10 to 15 years - Referenced as the current duration of the drug development process. Cost to bring a drug to market: $2 billion - Used as part of the R&D productivity problem framing; later Khan notes estimates can be even higher depending on modality. Clinical development spend: 70% - Khan says roughly 70% of drug-development dollars are spent in clinical development. Trial recruitment failures: 80% of studies fail to recruit on time - Cited as a major bottleneck AI can help address. Diversity/representation gap: majority of studies lack proper demographic representation - She notes many trials do not reflect the populations that will ultimately use the therapies. Enrollment improvement target: 40%-50% on-time enrollment - Khan says AI helped improve on-time recruitment from a poor baseline toward this range. J&J AI team growth: 5 people to 250 people - She describes scaling the internal AI organization at Johnson & Johnson. Recursion experiment throughput: 2.2 million experiments per week - Used to illustrate the scale of automated data generation supporting model training. Public/private biology understanding: 10%-15% understood - Khan says only a small fraction of biology is well understood today. Pipeline size: ~10 clinical-stage programs - Recursion’s current pipeline size as described by Khan. Pipeline mix: half oncology, half rare diseases - She describes the portfolio composition. Late discovery programs: 10+ programs - Number of programs beyond the clinic stage. Partnered programs: 10 more with 4 partners - She cites additional programs with external partners. RBM39 IND-enabling timeline: 18 months vs ~42 months industry average - She highlights speed from idea to IND-enabling studies for a first-in-class degrader. CDK7 exposure: order of magnitude higher - Early data showed PK exposure far above peers, matching design goals. Competition prize: >$500,000 - The sponsor mention for the TBXT challenge prize competition.

Pivotal Quotes: "Take the best of what industry has to offer and innovate on the rest." — Najat Khan: Her summary of Recursion’s approach to combining proven industry practices with AI-driven reinvention. "AI doesn't replace the researcher. The researcher is augmented by AI." — Najat Khan: Explaining why resistance to AI often stems from fear and misunderstanding. "Garbage in and garbage out." — Najat Khan: Her point that data quality and reproducibility are essential for reliable AI models in biotech.

Implications: The episode suggests AI will reshape biotech only through disciplined execution: better data, better teams, and fewer promises. For industry, the near-term wins are in trial design, recruitment, and hard-to-drug programs—not miracle drugs.