My notes from the Bits in Bio meetup last night.
Host: Pear VC
Event Page: https://lu.ma/e9jkzbp4?tk=mMivGO
Moderator: Ashton Teng (Kepler AI)
Panelists:
Kim Branson (GSK)
Imran Haque (Kimia Therapeutics)
Daniel Leventhal (Xaira Therapeutics)
Areas of R&D Most Impacted by AI
Kim Branson (GSK)
- Models guiding experiment selection.
- Predictive models for identifying clinical trial participants (reduce cost, improve targeting).
- Financial forecasting.
- Moving away from animal studies (mice) toward human subject selection.
- Need for faster model feedback cycles.
- Biggest gains expected in vaccines—AI can help select antigens.
Imran Haque (Kimia)
- AI useful where data is abundant and structured.
- “It’s called AI until it works—then it’s just computation.”
- Cell imaging under perturbations → phenotypic insights.
- Next-gen sequencing is costly vs. imaging.
- Scale: AI enables 30k new candidates per cycle vs. 3 historically.
- Challenge: deciding what to measure in the first place.
Challenges & Limitations
Kim Branson (GSK)
- More patient-level data available than ever → big opportunity for trial participant selection.
- Long timelines (e.g., 18-month studies) remain unchanged despite AI.
Imran Haque (Kimia)
- Automation in labs still underdeveloped; no one has end-to-end automated workflows.
Daniel Leventhal (Xaira)
- Core R&D competencies should not be outsourced.
Agentic & LLM Systems
Daniel Leventhal (Xaira)
- Early models were hard-coded; LLM orchestration enables modularity.
Kim Branson (GSK)
- “Automate the boring stuff.”
- Multi-agent frameworks:
- One agent speculates hypotheses.
- Another finds supporting data.
- Another scores it.
- Systems not replacing scientists but accelerating early-stage reasoning.
- No full end-to-end experimental autonomy yet.
Clinical Trials & Patient Selection
Kim Branson (GSK)
- First step: model which patients will respond (maybe using Quantitative Systems Pharmacology—QSP).
- Better patient population selection crucial for trial success.
Daniel Leventhal (Xaira)
- Patient population and target selection are key leverage points.
Partnerships & Industry Strategy
Kim Branson (GSK)
- Dislikes “top-down” intros from CEOs/board members—prefers value proven at lower levels.
- Will not onboard entire 14k-person R&D org to your product.
- Advice to startups:
- Partner with smaller players first.
- Then scale to big pharma for better deal terms.
- Tech stack compatibility required (Azure, AWS, GCP).
Audience Q&A Highlights
-
Lab automation & agents:
- Kim: Agents must interface with robots for lab workflows.
-
Strategic roadmaps:
- Clinical foundation models emerging.
- Chemistry, Manufacturing & Controls (CMC) still an underdeveloped, avoided area.
Key Takeaways
- AI is most impactful in experiment design, clinical trial optimization, and biomedical knowledge management.
- Data-rich areas (imaging, sequencing, functional assays) are driving real advances; cost differentials matter (imaging cheaper than sequencing).
- Agentic systems are gaining traction but remain focused on augmenting scientists, not replacing them.
- Patient selection and vaccine development are viewed as high-value domains for AI.
- Partnerships succeed when startups can demonstrate tangible value quickly, especially with infrastructure compatibility.
- Despite progress, long clinical timelines and core R&D tasks remain constraints.