Episode 49: AI in Pharma / Biotech

In this episode, we explore the high-stakes reality of AI integration in the life sciences. We move past the speculative hype to examine the mechanical necessity of this shift, driven by soaring R&D costs and an explosion of biological data, and how the industry has transitioned from a "serendipity-based" model to one of predictive, generative biology.

We break down the leaders of this revolution, comparing the "TechBio" pioneers like Insilico and Recursion with the massive infrastructure bets currently being made by Big Pharma giants. From the rise of autonomous, "closed-loop" labs to significant improvements in target discovery and preclinical timelines, we analyze where the technology is delivering measurable wins and where the "dirty data" gap still holds the industry back.

Finally, we address the hard truths of this transition, including the recent wave of industry layoffs and the emergence of the "Scientific Translator" as the most vital role in the modern lab. We look ahead to a 2030 landscape defined by digital control arms and personalized medicine, offering essential lessons for any professional navigating the risks and rewards of an AI-integrated pipeline.

🎧 Listen now, stay curious, and don’t forget to subscribe for new episodes every Thursday!

https://linktr.ee/maneadkhin

#Biotech #AI #Pharma #LifeSciences #Podcast #PetriDishPerspectives

Transcript

Hello and welcome to Petri Dish Perspectives, the podcast where we geek out about science and the companies shaping the future of healthcare. I’m your host, Manead, and I’m a PhD scientist by training, biotech storyteller by choice. With every new episode released on Thursday, my goal is to deliver digestible pieces of information on healthcare companies under 30 mins. 

The honeymoon phase is over. We’ve moved past the "AI is a magic wand" era and into the "AI is a mechanical necessity" era. Here is the blueprint of what that looks like across the industry.

Quick disclaimer, I give full credit to the original articles cited in the references in the transcript!

Grab a coffee or tea, settle in, and let’s jump in!

1. The Trigger: Why Pharma Finally Hit the "Go" Button

The shift wasn't just about FOMO. It was a mechanical response to three systemic failures:

• The Innovation Gap: We were hitting a ceiling with "Small Molecule 1.0." The cost to develop a drug had ballooned to roughly $2.6B, while the success rate remained stuck at a dismal 90% failure in the clinic. The industry was effectively running out of "low-hanging fruit."
• The Multi-Omics Explosion: We reached a point where the data generated by a single patient's tumor sample—genomics, transcriptomics, and spatial proteomics—was too dense for a human team to synthesize. We needed AI to find the "signal" in the noise of millions of variables.
• The AlphaFold Moment: When protein structure prediction was "solved" computationally, it proved that AI could handle the three-dimensional complexity of biology, moving the field from descriptive (what happened?) to predictive (what will happen?).

2. The Pioneers: The "Original Three" vs. Big Pharma

The "TechBio" Deep Dive

• Insilico Medicine: They are currently the "gold standard" for generative chemistry. Their lead candidate, rentosertib (INS018_055) for IPF, is the first wholly AI-discovered and AI-designed molecule to reach Phase II trials. In March 2026, they announced that their Pharma.AI platform is now producing candidates for "undruggable" targets in oncology and immunology with a 95% hit rate in preclinical validation.
• Recursion Pharmaceuticals: Known for their "Phenomics" approach, Recursion uses massive automated microscopy to run millions of "experiments in a well." Following their 2024 merger with Exscientia, they’ve combined Recursion’s biology-first data factory with Exscientia’s chemistry-first precision platform.
  The Pivot: It hasn't been all smooth sailing. In early 2025, they famously "trimmed" three major programs (including CCM and NF2) to extend their cash runway into 2027. This was a sobering lesson: AI can find the drug, but the clinic still has the final vote.
• Exscientia: Before the merger, Exscientia was the first to put an AI-designed drug into human trials (DSP-1181). Their focus on patient-tissue testing (using actual patient samples rather than mouse models) is now being integrated into the Recursion stack to improve the translation from "dry lab" to human biology.

How Big Pharma is Integrating

Big Pharma has moved from "curious" to "infrastructure-heavy."

• Roche: Rather than just licensing software, they are building their own hardware. They recently acquired thousands of Nvidia H200 AI chips to build a centralized "AI Factory" in Europe and the US. Their goal is to run digital twins of their manufacturing lines to double production speed.
• Eli Lilly: They’ve gone even further, announcing a massive AI Factory in October 2025 equipped with 1,016 Blackwell Ultra GPUs. Lilly is using this to dominate the "Metabolic Moat," specifically predicting which GLP-1 analogs will have the fewest side effects before they even hit a petri dish.

3. What AI Use Looks Like NOW (March 2026)

We are currently in the "Agentic Era." AI is no longer just a chatbot; it’s an active participant in the lab.

• Vibe Coding for Bio: Lab scientists are now using "intent-driven" coding. You don't write Python; you tell the AI the "vibe" or logic of the experiment, and it generates the bespoke scripts for the liquid-handling robots in real-time.
• Autonomous Wet Labs: We are seeing the rise of "closed-loop" systems. A predictive model designs a protein, sends the instructions to an automated synthesizer, tests the result, and feeds the data back into the model to iterate—all while the human researcher is asleep.
• Scientific Translators: The most valuable people in the room aren't "pure" data scientists anymore. They are Scientific Translators—PhD researchers who can navigate the nuanced intersection of $PI3K\alpha$ mutations and machine learning loss functions.

4. Areas of Visible Improvement

• Target Identification: We’ve seen a 40% reduction in the time it takes to move from a biological hypothesis to a validated drug target.
• ADME/Tox Prediction: AI models are now significantly better at predicting whether a molecule will be toxic to the liver ($p < 0.05$ improvement over traditional assays), reducing the number of "dead on arrival" molecules that enter Phase I.
• Clinical Enrollment: AI is being used to screen EMR data to find "needle-in-a-haystack" patients for rare disease trials, such as Juvenile Polyposis Syndrome (JPS), cutting recruitment times by months.

5. The Gaps: Why We Haven't Reached "Singularity"

• The "Dirty Data" Problem: 55% of AI pilots still fail because the historical data is a mess. AI can't fix a poorly recorded experiment from 1998.
• Biological "Hallucination": AI can design a molecule that binds perfectly to a receptor on a computer screen but fails to reach the target in a living, breathing human with a complex immune system.
• Regulatory "Black Box": The FDA still requires "explainability." If we can't explain why the AI chose a specific molecule, we can't get it through a safety review.

6. The AI-Integrated Future (2030 and Beyond)

• Digital Control Arms: We will eventually see trials where the placebo group is entirely digital—a "Digital Twin" of the patient population—eliminating the ethical and logistical hurdles of traditional control groups.
• N-of-1 Drug Design: For ultra-rare diseases, we may see drugs designed specifically for a single patient's genetic sequence, manufactured in small-batch "desktop" bioreactors.
• Quantum Integration: As quantum computing matures, we will finally be able to simulate the full quantum chemistry of a drug-receptor interaction, moving from "prediction" to "perfect simulation."

7. The Human Element: Layoffs and the "Efficiency" Squeeze

Has AI caused layoffs? Yes, but it's a reallocation, not a total replacement.

• Structural Reductions: We saw 20-25% staff cuts at Exscientia and Recursion post-merger. These were largely in "traditional" R&D roles that were superseded by automated high-throughput pipelines.
• The "Middle Management" Crunch: AI is very good at doing the work of a junior analyst or a lab tech. This has led to a "hollowing out" of entry-level positions in some biotechs, as one "Scientific Translator" can now do the work of five people.

8. Lessons & Next Steps for the Industry

1. Stop "AI Washing": Investors now look for clinical proof, not just "AI platforms." The Novartis deal for SNV4818 proves that $2B upfront only happens when the science is mutant-selective and clinically sound.
2. Own Your Infrastructure: If you rely entirely on third-party AI, you are a customer, not a leader. The "Roche/Lilly" model of building internal AI Factories is the new standard for Big Pharma.
3. Data Governance is the Product: Your R&D is only as good as your metadata. Clean, annotated, prospective data is the most valuable asset you own.

As we close the door on the "Autonomous Lab" for today, one thing is certain: the era of AI-Washing is dead. In its place, we have a gritty, high-stakes reality where billion-dollar deals like the Novartis-Synnovation pact are won or lost on the back of data integrity.

We’ve moved from asking "Can AI find a drug?" to "How fast can we manufacture it, and how cleanly can we explain the mechanism to the FDA?" Whether you’re working in the trenches of oncology or navigating the complex intellectual property landscape of a startup, the "black box" is no longer an excuse. The future belongs to the Scientific Translators—those who can keep one foot in the wet lab and the other in the GPU cluster.

Biology has always been a game of pattern recognition; we just finally have the processing power to see the full picture.

This has been Petri Dish Perspectives. I’m Manead. Thanks for listening. See you next Thursday. Good bye.