💊 The Most Expensive Mistake in Drug Discovery

Imagine this:

A company spends:

• 4 years
• $150M
• dozens of scientists

developing a drug against Target X.

✅The molecule works exactly as intended.

✅The chemistry is solid.

✅The safety profile looks acceptable.

❌ And then the clinical trial fails.

Why?

Because Target X was never a major driver of the disease in the first place.

The painful part is that this isn't a chemistry failure.

It's a biology failure.

One pattern that keeps appearing in post-mortems of failed programs is that the underlying biological hypothesis was weaker than people thought.

🎯 Not every failure is caused by poor target selection, of course.

But target selection may be one of the most expensive decisions made in the entire R&D process.

❔ Which raises a question:

How much evidence is enough before we decide a target is worth pursuing?

Human genetics?
Clinical data?
Animal models?
Literature evidence?
...
Or some combination of all of them?

I'd love to hear how your team approaches this!👇🏽

🔥Target Intelligence Digest 🔥

The Signals Shaping Drug Discovery Right Now

🟣 1. Isomorphic Labs Raises $2.1B
DeepMind-backed Isomorphic Labs has raised $2.1 billion to scale its AI-driven drug discovery platform built around AlphaFold and computational biology.

🔗 https://www.reuters.com/legal/litigation/google-backed-isomorphic-raises-21-billion-scale-ai-driven-drug-discovery-2026-05-12/

Why it matters

Most headlines focus on AI.
The more interesting signal is where investors are placing their bets.
This isn't just about generating molecules faster.
It's about understanding biology better — mapping disease mechanisms, protein interactions, and therapeutic opportunities before entering the costly stages of development.

🟣 2. Eli Lilly Doubles Down on AI Drug Discovery

Eli Lilly expanded its collaboration with Insilico Medicine in a deal worth up to $2.75 billion, gaining access to AI-enabled discovery capabilities and preclinical assets.

🔗 https://www.reuters.com/business/healthcare-pharmaceuticals/eli-lilly-extends-partnership-with-insilico-medicine-ai-powered-drug-discovery-2026-03-30/

Why it matters

A few years ago, AI companies mostly sold software.
Today, large pharma companies are increasingly buying:
discovery platforms
target identification capabilities
biological insight
decision-making infrastructure
That feels like a significant shift in how value is being created.

🟣 3. Takeda Signs $1.7B AI Partnership

Takeda entered a partnership with Iambic Therapeutics worth more than $1.7 billion to discover new therapies for oncology and gastrointestinal diseases.

🔗 https://www.reuters.com/business/healthcare-pharmaceuticals/takeda-deepens-ai-drug-discovery-push-with-17-billion-iambic-deal-2026-02-09/

Signal

Another major pharmaceutical company is making a large bet on:
AI-enabled target discovery
protein-ligand interaction prediction
computational biology
At this point, these are no longer isolated experiments.
They're becoming part of the industry's core strategy.

🟣 4. Amazon Enters AI Biology

Amazon recently introduced Amazon Bio Discovery, a platform combining more than 40 AI biology models to help researchers prioritize candidates for experimental validation.

🔗 https://www.techradar.com/pro/amazons-new-ai-bio-discovery-tool-can-provide-every-researcher-with-lab-in-the-loop-drug-discovery-40-ai-biology-models-can-filter-300-000-novel-antibody-candidates-down-to-the-top-results-for-testing-in-just-weeks

Why it matters

Another signal that AI biology is becoming infrastructure.
Not a niche capability.
Not a startup experiment.
Infrastructure.

🟣 5. A New Category Is Emerging: AI for Target Selection

Insilico continues to expand platforms such as TargetPro and TargetBench, focusing on target discovery, validation, and druggability assessment.

🔗 https://www.aastocks.com/en/stocks/news/aafn-con/NOW.1518694/ipo-news/AAFN

Why it matters

This reflects a broader shift happening across the industry.
The old question:
Can we make the molecule?
The new question:
Should we pursue this target at all?

🧠 Research Intelligence Signal

A recent study comparing a specialized pharma intelligence platform against leading general-purpose LLMs found that the domain-specific system significantly outperformed frontier models in identifying and analyzing pharmaceutical assets and pipelines.

🔗 https://arxiv.org/abs/2605.04908

Why it matters

This is particularly interesting for anyone building AI tools for life sciences.
The lesson is becoming increasingly clear:
The competitive advantage isn't just the model.
It's the combination of:
curated data
domain-specific workflows
structured biological knowledge

🏹 My Take

Looking across these developments, I keep seeing the same pattern.

For years, the industry was obsessed with Molecule Intelligence:
Can AI generate better molecules?

Now a new priority is emerging:
Can AI help us understand the biology well enough to make better decisions before we spend hundreds of millions of dollars? 

If this trend continues, some of the most valuable companies of the next decade may not be the ones generating molecules.

They may be the ones helping researchers decide which biology is worth pursuing in the first place. 🚀

🚨 Target Intelligence Digest
Signals from the Last Two Weeks

🔄 Signal #1: Big Pharma Is Buying Research Infrastructure, Not Just Drugs

Merck KGaA announced its acquisition of Bio-Techne for $11.3B. Bio-Techne isn't known for a single blockbuster drug. Instead, it provides research tools, proteins, antibodies, reagents, and scientific infrastructure used across life sciences.

🔗 https://www.reuters.com/business/healthcare-pharmaceuticals/germanys-merck-buy-bio-techne-73-per-share-2026-06-25/

Why it matters

This isn't a bet on one therapeutic asset.
It's a bet on the infrastructure that generates biological knowledge.

🔄 Signal #2: Human Genetics Keeps Winning

A large-scale analysis published recently examined more than 26,000 target-disease pairs.
The key finding:
Targets supported by human genetic evidence were approximately 3x more likely to become approved drugs than targets without genetic support.

🔗 https://arxiv.org/abs/2606.14823

Why it matters

For years, genetics was viewed as one signal among many.
Increasingly, it looks like one of the strongest predictors of downstream clinical success.
This helps explain why so many companies continue investing heavily in:
target validation
genetics platforms
disease mechanism discovery
causal biology
rather than focusing exclusively on chemistry.

🔄 Signal #3: AI Deals Are Getting Bigger — But the Focus Is Changing

Over the past few weeks:
Alnylam and Inceptive announced a collaboration worth up to $2B
Insilico Medicine and SK Biopharmaceuticals announced a collaboration worth up to $2.5B
Merck and Protillion announced a partnership worth up to $510M

🔗 https://www.nasdaq.com/press-release/alnylam-and-inceptive-form-strategic-ai-collaboration-accelerate-discovery-rnai
🔗 https://www.pharmexec.com/view/insilico-collaboration-sk-biopharmaceutical-ai-powered-cns-treatment
🔗 https://www.fiercebiotech.com/biotech/merck-inks-510m-biobucks-data-generation-partnership-protillion

Interesting observation:

A few years ago, the dominant pitch was:
AI can generate molecules.
Today, the conversation increasingly sounds like:
AI can identify targets, generate evidence, design experiments, and reduce biological uncertainty.

🔄 Signal #4: AI Drug Discovery Is Starting to Deliver More Clinical Evidence

Insilico Medicine recently announced first-in-human dosing for its AI-developed NLRP3 inhibitor program.

🔗 https://www.marketscreener.com/news/insilico-completes-first-in-human-dosing-in-phase-i-clinical-study-of-ai-driven-nlrp3-inhibitor-ism8-ce7f5cdcdf81fe25

Why it matters

For years, the industry heard:
AI drug discovery is promising.
Now the conversation is shifting toward:
Show us clinical outcomes.
The next few years will likely determine which AI drug discovery companies become true biopharma players and which remain software stories.

🔄 Signal #5: Pharma M&A Is Moving Earlier

According to recent industry analysis, approximately 60% of biotech M&A value now comes from Phase II or earlier assets, a significant increase compared to historical norms.

🔗 https://www.ft.com/content/9e79e41b-707b-47ed-9c75-1779a7978f68

Why it matters

Large pharmaceutical companies appear increasingly willing to acquire programs earlier in development.
Why?
Because once a biological hypothesis is strongly validated, the asset often becomes dramatically more expensive.
In other words:
Confidence in the biology may now be worth paying for.

🧠 My Take:

The most interesting pattern I'm seeing across all of these signals is that they point toward the same fundamental question:
How do we increase confidence before committing years of work and hundreds of millions of dollars?

Because the industry's biggest challenge may no longer be making drugs.
It may be deciding which biology deserves a drug in the first place. 🚀

🚨 Anthropic just changed the game for scientific research.
Yesterday, Anthropic launched Claude Science — an AI workspace built specifically for scientists.

It can already:

• review thousands of papers
• analyze genomics and single-cell datasets
• design molecules
• connect to PubMed, UniProt, PDB, ChEMBL and many other scientific resources
• orchestrate multi-step research workflows
• generate publication-ready figures and manuscripts

This is a huge step forward.
For many researchers, "using AI" will no longer mean switching between ten different tools.
It will mean opening Claude.
So...

Does that mean the problem of drug discovery is solved?

Not even close.

Because after Claude summarizes 5,000 papers, one expensive question still remains:
😵 Would you invest $50M into this target?

And that's where today's AI still struggles.
➕It can retrieve information.
➕It can summarize evidence.
➕It can even generate hypotheses.

But it still doesn't tell you:
• How strong is the biological evidence?
• Which findings are contradictory?
• Which experiments deserve more weight?
• Where is the uncertainty?
• Is this biology actually robust enough to justify a drug program?

Those aren't literature questions.
They're decision questions.

I think this is where the next generation of AI for life sciences is heading.
Not toward replacing scientists.
Toward helping them make higher-confidence decisions.

🔄 That's exactly the direction I'm currently exploring.
If you're interested in Target Intelligence, Research Intelligence, and how AI can support real R&D decisions rather than just summarize papers, I'd love to connect.

🗂 Inside My Research #1

I counted where biotech founders actually spend their time before choosing a target.

Over the last few weeks, I've been mapping what happens before a biotech company decides:
"Yes, let's invest in this target."

🤨 What surprised me wasn't the science.
It was the number of disconnected decisions.

A typical workflow looks something like this:

• PubMed → Is there enough published biology?
• Open Targets → Is there human genetic evidence?
• ClinicalTrials → Has anyone already failed?
• ChEMBL → Are there known compounds?
• Company websites → Who else is working on this?
• Patents → Is there still room to operate?
• Internal slide deck → Can we convince investors?

None of these questions are difficult individually.

The problem is that they live in completely different worlds.
Every scientist I talk to has developed their own way of stitching these sources together.

Usually it's browser tabs.
Bookmarks.
Spreadsheets.
A personal collection of scripts.
That's not really a workflow.
It's institutional memory living inside one person's head. 🧑‍🔬

😵 I think this is where AI becomes genuinely interesting.
Not because it can summarize another paper. ☝🏽️

But because it can finally connect pieces that were never designed to work together...

💎 Building AI for Decision Intelligence in drug discovery.

For the past eight years, I've worked inside biotech startups.
My role was never to do the science itself.

🧪It was to help scientific teams operate:
coordinating projects, working with customers, preparing documentation, supporting grant applications, and keeping complex R&D programs moving.

Working so closely with researchers gave me a front-row seat to one recurring problem:
Critical information rarely lives in one place.
It's scattered across publications, databases, clinical evidence, patents, spreadsheets, and internal knowledge.

💊Most drug discovery data already exists.
The hard part is turning it into a decision.
That's the problem I'm exploring.


🖱What I'm building now:

AI-assisted Target Due Diligence — structured evidence assessment that helps biotech founders and CSOs answer that question before committing years of work and millions in R&D.

Not just data.
Not just summaries.
A structured risk assessment: what the biology actually supports, where the evidence contradicts itself, and why.

📣 I'm currently speaking with founders, CSOs, Heads of Research, and scientists to better understand how these decisions are made in the real world.
If that's your world, I'd genuinely love to hear how you and your team evaluate new targets.

➡️In return, you'll get early access to what I'm building, opportunities to shape the product, and a chance to try new features before they're publicly released.⬅️


🤝 Feel free to DM me.

⭕️ Channel of a project
🎰My LinkedIn

🤔 I Changed My Mind #1
I used to think biotech's biggest bottleneck was finding proper information.

Now I think it's something much more subtle.

It's confidence.

Think about it.
Every week there are:

🌀thousands of new papers;
🌀new preprints;
🌀new clinical trial updates;
🌀partnership announcements;
🌀patents;
🌀conference abstracts.

Nobody can read everything.
So researchers don't actually optimize for completeness.
They optimize for confidence.

The real question becomes:
❓"Have I seen enough evidence to make a decision?"

That's a very different problem from literature search.

And I suspect that's why so many AI tools struggle to become part of everyday research.

They're optimized to answer questions.
Researchers are trying to reduce uncertainty.
Those aren't the same thing.☝🏽️

One gives information.
The other changes decisions.
I think the next generation of research tools will be judged not by how many papers they summarize…

…but by whether they help teams commit to (or reject) a biological hypothesis faster.

What do you think? 👇🏽

💵 Expensive Question #1

How many failed clinical trials would make you abandon a target?

Three?
Five?
Ten?
There isn't a correct answer.
But here's something interesting.

Most teams don't just count failures.
They ask:

Why did it fail?
Was the target wrong?
Was the molecule wrong?
Was the patient population wrong?
Was the endpoint unrealistic?
Was the biomarker poorly chosen?

A failed trial isn't one piece of evidence.
It's five different signals.
And that changes how you evaluate the next program built around the same biology.

This is one reason why target evaluation still takes experienced scientists days or weeks.

The information exists.
The reasoning doesn't...


That's why I'm doing it
👇🏽
https://t.me/targetduediligence

😵 AI Drug Discovery Weekly
The biggest signals from the past week.

🚀 1. Anthropic launches Claude Science (Beta)

This is the biggest story of the week.
Claude Science is no longer just an AI chatbot—it has become a dedicated scientific workspace. Researchers can now search literature, analyze genomics, run computational workflows, generate publication-ready figures, and draft manuscripts from a single interface with built-in reproducibility and reviewer agents. (TechRadar)

👉 Why it matters: AI companies are shifting from building general models to building complete operating systems for scientific research.

🔗 Read more:
TechRadar coverage
The Verge analysis

💊 2. Anthropic says it wants to develop its own drugs

Perhaps the most surprising announcement.
Anthropic confirmed that it intends not only to build AI tools for scientists, but also to pursue its own drug discovery efforts, initially focusing on neglected diseases. (The Verge)

👉 Signal: Frontier AI labs are beginning to act like biotech companies—not just software vendors.

🔗 Read more:
STAT News
The Verge

🤝 3. Takeda signs up to a $600M AI partnership with Insilico Medicine

Takeda announced a new collaboration with Insilico Medicine worth up to $600 million to accelerate AI-driven target identification and drug discovery. (The Wall Street Journal)

👉 Signal: Big Pharma continues to increase investment in AI-native drug discovery platforms.

🔗 Read more:
Wall Street Journal coverage

💰 4. Anthropic opens its AI for Science grant program

Anthropic is offering up to $30,000 in compute credits for selected scientific research projects through its new AI for Science initiative. The program is designed to support researchers using Claude Science for biology, chemistry, and related fields. (TechRadar)

👉 Signal: AI companies are actively investing in growing the scientific AI ecosystem, not just selling subscriptions.

🔗 Read more:
TechRadar coverage



💬 My takeaway

The conversation is changing.
Just a year ago, most companies competed on who could generate better molecules.
Today, the race is becoming who can help scientists make better decisions.

That shift is exactly why I'm spending so much time thinking about AI-assisted Scientific Due Diligence rather than another literature search tool. The next generation of AI in drug discovery won't just generate more data—it will help decide what is actually worth pursuing.