Unlearn.AI

See you at SCOPE X in a week! Don't miss our VP of Product Kwame Marfo's presentation, "The Trust Dividend: How Regulatory-Grade AI Compounds Value Across the Trial Lifecycle" on May 19 at 10am. Every #clinicaldevelopment team will adopt #AI — the harder question is which AI is trustworthy enough to drive key decisions. Kwame will be drawing on Unlearn's journey to regulatory qualification to show how trust earned once compounds across planning, monitoring, and analysis. 🚀 ——— #SCOPEsummit

FOLFIRINOX vs. gem + nab-paclitaxel is one of the most common first-line decisions in advanced pancreatic cancer — but no RCT has ever directly compared them. Indirect methods like NMA and target trial emulation each have limitations: transitivity assumptions break down when trial populations differ, and RWD-based approaches are resource-intensive and frequently study a different population than the trials in question. In Part 4️⃣ of our #oncology trial design series, we show how trial-calibrated generative modeling offers a third path—producing patient-level simulations that respect published RCT results while adjusting for baseline covariate imbalances. Applied to PRODIGE4 vs. MPACT, the approach softens the apparent 1-year RMST advantage of FOLFIRINOX by about half. Read the full post: na2.hubs.ly/H05lFs50

We're heading to #SCOPEsummit in Boston later this month, and the conversations happening there are ones we think about every day. How do we design smarter #trials, increase confidence, and build #AI that's trustworthy enough to drive key decisions across #clinicaldevelopment? Our VP of Product Kwame Marfo will be addressing exactly that on May 19 at 10am: The Trust Dividend: How Regulatory-Grade AI Compounds Value Across the Trial Lifecycle. See you there! 🚀

May is #ALSAwarenessMonth. #ALS is relentlessly progressive, with no cure and painfully limited treatment options — making confident, efficient #clinicaldevelopment not just valuable, but essential. We're proud to partner with ProJenX, QurAlis, Trace Neuroscience, SOLA Biosciences, VectorY Therapeutics, and others at the forefront of ALS #drugdevelopment helping them reach go/no-go decisions with greater confidence — so promising therapies can move forward faster for people who are waiting. ——— @alsassociation

#UnlearnerSpotlight 🚀 "I'm surrounded by sharp, thoughtful people who care about the work and are very supportive of each other, all working toward smarter trials and faster treatments for patients.”

We predicted the BREAKWATER control arm without ever training on BREAKWATER data. That's the kind of question most #oncology teams want to answer during the design phase but can't, because the work to run each scenario could take months. Part 3️⃣ of our oncology series is about what changes when that's no longer the constraint. When you can ask "what does the control arm look like under current standard of care?" and get a calibrated answer before the protocol is finalized. ❓ What would your team do differently with that kind of flexibility during the design phase? na2.hubs.ly/H04VWh80

🎉 We're proud to announce that Unlearn has been named the winner of the 2026 Fierce Biotech Outsourcing Award for AI & Advanced Analytics. This recognition is a testament to our team's dedication to transforming how #clinicaltrials are designed, analyzed, and advanced — empowering sponsors to make faster, more confident decisions that get treatments to patients sooner. We also want to recognize the finalists in our category — the depth of innovation across this field speaks to the incredible momentum #AI is bringing to #drugdevelopment. 🚀 Read more: na2.hubs.ly/H04TD370

Unlearn team at #DISS2026! 🎉 Last week our team had a great time at the Duke Industry Statistics Symposium, exploring the latest in AI/ML and data innovation in pharmaceutical development — from adaptive trial designs to real-world evidence and beyond. Grateful for the insightful conversations and connections! ——— @DukeBiostats

Behind every #clinicaltrial is a person waiting for a breakthrough. That's why we use AI-generated #digitaltwins to make PD trials more efficient, reducing sample sizes by up to 23% and control arms by up to 38%, helping promising therapies reach patients faster. This #ParkinsonsAwarenessMonth, see what that looks like it our latest PD case study ➡️ na2.hubs.ly/H04NKd80

5️⃣ patients. That's how many in our available datasets fully matched the eligibility criteria for a recent BRAF-mutant colorectal #cancer trial. A data-matching approach stops there. Part 2 of our #oncology series explains what a modeling approach makes possible when the #data runs out, including how we predicted that trial's control arm overall survival without ever training on its data. na2.hubs.ly/H04Frjy0

Q1 was a big quarter for Unlearn. ➡️ We launched TrialPioneer — a unified workspace that helps sponsors move from scattered planning to evidence-grounded #trialdesign, before a protocol is ever finalized. ➡️ We deepened our digital twin models in #Huntingtons disease, announced new #ALS partnerships with VectorY Therapeutics and SOLA Biosciences, and introduced a pan-cancer foundation model trained on ~300,000 tumor biopsies. 🚀 Our CEO Steve Herne breaks down what we built, why it matters, and where we're headed. Read the full blog post: na2.hubs.ly/H04Dx_J0

A recurring challenge in precision #oncology: as treatments are tested on narrower populations, patient-level data alone becomes insufficient to provide the insight needed to understand outcomes, and the data that does exist is costly and time-consuming to collect. In our new whitepaper, we present a case that illustrates this directly. For the BREAKWATER trial in BRAF V600E mutant mCRC, only 5 patients in our source data matched the eligibility criteria. A direct data-matching approach would stop there. Our foundation model generalized from a broader pool of nearly 800 patients with BRAF V600E mutations across indications and regimens, and was calibrated to prior unselected mCRC trials for this prediction. It produced OS predictions that closely matched BREAKWATER's published results, without ever being calibrated to that trial directly. This is the practical value of combining patient-level data with published trial results. The model learns from related populations and indications, transfers that knowledge to narrow cohorts, and anchors its predictions to established trial evidence. It does not require a perfectly matched dataset to produce meaningful estimates. The same capability underpins precision trial simulation, synthetic comparator generation, and comparative effectiveness analyses. Leading sponsors are applying it today in their oncology pipelines. 📄 Download the whitepaper for the full methodology and case studies, or reach out directly if you'd like to learn more. na2.hubs.ly/H04B5C-0

The lung cancer community came together in Copenhagen last week for #ELCC26, and it was a week of real conversations about advancing #clinicaldevelopment in #oncology. As populations narrow — tighter eligibility, evolving standards of care — benchmarking expected outcomes gets harder. Patient-level data alone isn't enough, and collecting new RWD for every subgroup is costly with limited yield. We presented validation results for a foundation model in oncology that generates patient-level outcome predictions calibrated to published trial results — no matched dataset required for each population of interest. 🚀

💊 #Clinicaltrial design sits between evidence discovery and protocol execution, a middle space that has never had a dedicated home. Most organizations fill that gap with tools that weren’t built for it. They work well enough for one iteration. As designs evolve, version control becomes manual, provenance gets lost between documents, and scenario exploration stays serialized. The result: fewer design options are rigorously stress-tested than the team would’ve liked, because running one more scenario wasn't worth the delay. 📄 Our new paper puts a number on the cost. Download it here: na2.hubs.ly/H04w_Lc0

🎗️ Standard of care in #oncology evolves rapidly, and treatments are increasingly tested on narrower, biomarker-defined populations. As this happens, the ability of patient-level data alone to provide the insight needed to understand outcomes diminishes. RWD sources are also costly and time-consuming to collect, especially for narrowly defined populations. Published trial results offer population-level authority, but cannot be queried at the patient level or transferred across eligibility criteria. The ability to combine these two sources, harnessing the strength of each, would be of tremendous practical value. Today we are publishing a whitepaper describing how we approach this problem. Our approach, FRESH modeling (Fusion of Recent Evidence and Subject Histories), pairs a pan-cancer foundation model trained on biopsy assays from approximately 300,000 cancer patients with a calibration procedure that anchors patient-level predictions to published trial results. The result is a system that integrates patient-level granularity with population-level rigor, without requiring a bespoke data acquisition for each new cohort or #trialdesign question. Practical applications include precision trial simulation, calibrated synthetic comparator generation, and head-to-head comparative effectiveness analyses. Leading sponsors are already working with us to apply this across their oncology programs. 📄 The whitepaper includes validation case studies across NSCLC and mCRC. Download below, or reach out directly if you'd like to discuss how this applies to your pipeline. na2.hubs.ly/H04w5qq0

The CNS community came together in Copenhagen for #ADPD2026 and it was a week of real conversations about where #clinicaldevelopment needs to go next. We shared new work on how #digitaltwins of trial participants can improve trial design in Alzheimer’s and Parkinson’s, including a simulation of TRAILBLAZER-ALZ 2 showing up to a 15% increase in power across key endpoints, without increasing sample size. Unlearn’s Co-founder and Chief Scientific Officer, Jon Walsh, also joined Forum 2 to discuss how our AI clinical trial solutions are being applied in practice in AD/PD and where the field is heading. More than anything, this week was about being in the room: discussing challenges and connecting with the people driving this work forward. Appreciative of the conversations and excited about what comes next. 🚀 ——— @adpdnet

#ICYMI: 🎉 Unlearn is excited to partner with SOLA Biosciences on an early-phase ALS study of SOL-257, supported by our AI-generated #digitaltwins. We're grateful for the opportunity to help advance clinical research for people living with #ALS. Read the full announcement: na2.hubs.ly/H04kxv_0

Every iteration of a protocol before finalization has a cost: evidence refresh, assumption reconciliation, and cross-functional realignment. For study teams, it's felt as time. For portfolio leaders, it has a dollar equivalent that compounds across every avoidable loop, before a single patient enrolls. We put a number on it. na2.hubs.ly/H04kwYy0

🎯 Predicting trial outcomes is getting harder, and the tools most teams rely on weren’t built for this. As patient populations narrow and standards of care shift faster than trials can read out, the assumptions underlying #trialdesign are increasingly built on shaky ground. Part 1 of our 2-part series breaks down why outcome prediction has become the hardest problem in #oncology #drugdevelopment, and why solving it starts with combining the granularity of patient-level data with the reliability of published trial evidence. Read it here: na2.hubs.ly/H04mg_z0

#UnlearnerSpotlight 🚀 “My favorite part is the mission, and the intentionality behind how we pursue it. We're not racing to layer AI onto everything, we're focused on building trust first. That's what excites me: working somewhere that believes in meeting people where they are, respecting their rigor, and bringing them along.”