Insilico Medicine: Inside the Most Validated AI-Drug-Discovery Platform in Mid-2026

Insilico Medicine is the most clinically advanced generative AI drug-discovery company operating today. Founded in 2014 by Alex Zhavoronkov, Ph.D., it is now a publicly listed clinical-stage biotech with dual headquarters in Boston, Massachusetts and Hong Kong, additional sites in New York, Montreal, Shanghai, Suzhou, Taipei, Yixing, and Abu Dhabi, a 2024 employee headcount of roughly 349 (about 78% in R&D), and a pipeline of more than 30 programs built entirely on its proprietary Pharma.AI platform. The most-watched program in the field — the TNIK inhibitor rentosertib (formerly INS018_055 / ISM001-055) for idiopathic pulmonary fibrosis — became the first drug with both its biological target and its chemical structure generated end-to-end by AI to produce positive mid-stage human efficacy data when the GENESIS-IPF Phase 2a results were published in Nature Medicine on June 3, 2025. That clinical proof, combined with the December 30, 2025 Hong Kong Stock Exchange listing (HKEX: 3696) and a Phase 3 initiation on July 7, 2026, has moved Insilico from “thesis-stage AI-bio company” into the rare tier of AI-drug platforms that the FDA, the China CDE, and thirteen of the world’s top twenty pharma companies treat as a real counterparty.

What follows is a source-by-source read on what Insilico has actually built, what rentosertib’s clinical record actually says, and what the marketing-grade writeups in circulation right now are getting wrong.

The Pharma.AI Platform — What Each Engine Actually Does

Insilico’s flagship product is Pharma.AI, a four-pillar software platform that maps onto the four stages of preclinical drug discovery:

• Biology42 / PandaOmics for target discovery and biomarker identification. The peer-reviewed reference is Kamya et al., Journal of Chemical Information and Modeling (February 25, 2024, DOI 10.1021/acs.jcim.3c01619). The engine uses more than 20 AI models over 1.3M disease-specific omics samples, 47M publications, 998K clinical-trial records, 5.5M patents, 3.2M grants, and 15K clinical-stage compounds, scoring targets across 20+ axes (novelty, druggability, safety, aging-association, commercial tractability). The pathway analysis uses the proprietary iPANDA algorithm (Ozerov et al., Nature Communications 7:13427, 2016).
• Chemistry42 for de novo small-molecule generation. The peer-reviewed reference is Ivanenkov et al., J. Chem. Inf. Model. 63:695–701 (February 2, 2023, DOI 10.1021/acs.jcim.2c01191). Chemistry42 runs an ensemble of 40+ generative models in parallel (the public number has drifted upward from 30+ in older marketing; the JCIM paper settles on 40+), including GANs (the company was one of the first to publish a deep adversarial model for chemistry in 2016), VAEs, flow models, transformers, and the proprietary GENTRL algorithm that produced a DDR1 inhibitor in 21 days in a 2019 Nature Biotechnology paper. Hits are scored through 2D and 3D reward modules and fed back to the generators through a multi-agent reinforcement-learning loop. The published sub-modules — Alchemistry (relative binding free energy), Golden Cubes (kinome selectivity via multi-dimensional self-organizing maps), ADMET Profiling, and Retrosynthesis — are documented on Insilico’s product page.
• Medicine42 / inClinico for clinical trial prediction. The peer-reviewed reference is Aliper et al., Clinical Pharmacology & Therapeutics 114:972–980 (2023, DOI 10.1002/cpt.3008). The published prospective accuracy on Phase 2 to Phase 3 transition is 79% over a training set of 55,600+ unique Phase 2 trials from the prior seven years. The product is currently sold to buy-side analysts, BD&L teams, and clinical researchers (per Insilico’s product page).
• Science42 for AI-augmented research automation. The umbrella product here is DORA (Draft Outline Research Assistant), a multi-agent LLM research tool that drafts scientific articles, grant applications, and review papers from a library of 15+ templates. DORA Community Edition was open-sourced on GitHub (insilicomedicine/DORA) on January 8, 2026 under Apache 2.0, supported by Microsoft Azure. The Science42 stack also includes PandaClaw and LabClaw (announced May 6, 2026), the latter an “Agent-Guard” laboratory operating system with 5 collaborative AI agents and 28 specialized skill modules that orchestrate Insilico’s Life Star 2 automated lab.

Two additional products round out the platform. Nach01 is a multimodal “natural and chemical language” foundation model launched on AWS Marketplace on June 10, 2025 and integrated with Microsoft Discovery; it is the successor to nach0, published in Chemical Science 15:8380–8389 (2024, DOI 10.1039/d4sc00966e), which was the first multimodal LLM to combine biomedical text and SMILES in a single encoder-decoder trained with NVIDIA NeMo. Precious3GPT is a multi-omics, multi-species, multi-tissue transformer for aging research, open-sourced on GitHub and Hugging Face; the bioRxiv preprint is dated July 2024. Science MMAI Gym, announced January 2026, is the training and benchmarking framework under which Insilico co-developed LFM2-2.6B-MMAI with Liquid AI (presented at ICLR 2026) and a longevity-specific foundation model with Human Longevity. The MMAI Gym is positioned as a “specialized pharmaceutical trainer” that upgrades general-purpose LLMs into domain-specific drug-discovery models — a one-line summary of the strategy is “on-premise small models beat 27B-parameter generalists on ADMET tasks.”

The platform is genuinely end-to-end in a way that most AI-bio companies are not. The 2019 Nature Biotechnology GENTRL paper (DDR1, 21 days from target to molecule) and the 2024 Nature Biotechnology TNIK paper (full pre-clinical-to-Phase 1 path on rentosertib) are the two foundational peer-reviewed demonstrations. The 2025 Nature Medicine paper (rento­sertib Phase 2a) is the third, and the one that actually matters for the regulatory question.

Rentosertib — The Clinical Proof

Rentosertib is a small-molecule inhibitor of TNIK (TRAF2- and NCK-interacting kinase), an intracellular serine/threonine kinase implicated in fibrosis-driving and inflammation-related pathways (Wnt, TGF-β, Hippo/YAP-TAZ, JNK, NF-κB). The target was prioritized by PandaOmics and the molecule was generated and optimized by Chemistry42; the discovery-to-Phase 1 story was published in Ren et al., Nature Biotechnology (DOI 10.1038/s41587-024-02143-0, published online March 8, 2024, print issue January 2025). The full pre-clinical-to-Phase 1 workup — including anti-fibrotic activity across lung, kidney, and skin in animal models, plus a Phase 1 trial in 78 healthy participants (NCT05154240) and a parallel Chinese Phase 1 (CTR20221542) — was completed in 18 months from target identification, with fewer than 80 molecules synthesized. The FDA granted Orphan Drug Designation in February 2023, the China CDE granted Breakthrough Therapy Designation in May 2025, and an IND for the U.S. Phase 2a cleared in June 2023.

The Phase 2a GENESIS-IPF trial (NCT05938920) enrolled 71 IPF patients across 21 sites in China, randomized 1:1:1:1 to placebo, 30 mg QD rentosertib, 30 mg BID rentosertib, or 60 mg QD rentosertib, treated for 12 weeks on top of standard-of-care antifibrotic therapy. The primary endpoint was the percentage of participants with at least one treatment-emergent adverse event (TEAE) — a safety endpoint, not FVC. That distinction matters: the Phase 2a was a safety trial, with FVC as a secondary efficacy endpoint, and the trial was not powered for FVC. The published Nature Medicine paper (Xu, Ren, Wang et al., Nat. Med. 31:2602–2610, June 3, 2025, DOI 10.1038/s41591-025-03743-2, PMID 40461817) reports the following:

• Primary safety endpoint met. TEAE rates were similar across arms: 72.2% / 83.3% / 83.3% / 70.6% (placebo). Treatment-related SAEs were rare and balanced. The most common TEAEs leading to discontinuation were liver-related events and diarrhea.
• FVC secondary endpoint showed a dose-dependent signal. 60 mg QD arm +98.4 mL (95% CI 10.9 to 185.9) vs placebo -20.3 mL (95% CI -116.1 to 75.6). The 30 mg QD arm was -27.0 mL (similar to placebo); the 30 mg BID arm was +19.7 mL (CI crosses zero). The 95% CI on placebo (-116.1 to +75.6) crosses zero — the placebo arm is consistent with both slight decline and slight improvement.
• Liver toxicity is a real signal. 7 of 55 active-arm patients (12.7%) discontinued due to liver injury, with 4 of 7 on concurrent nintedanib. The 30 mg BID arm had 22.2% discontinuation for liver injury; the 60 mg QD arm had 17%. One patient in the 30 mg BID arm died of heart failure deemed unrelated to treatment.
• FVC gain was concentrated in patients not on background antifibrotic therapy. In patients receiving 60 mg QD without nintedanib or pirfenidone, FVC improved +187.8 mL (CI 68.6 to 306.9). In patients on background antifibrotic therapy, the FVC signal disappeared — a drug-drug interaction that the authors flag as needing further study.
• Most other secondary endpoints were not significant. DLCO, FEV1, 6MWD were not different across arms. Cough-related QOL (LCQ) was significantly improved in the 60 mg arm (P=0.0495, two-sided). Three acute IPF exacerbations in the 60 mg arm vs one in placebo, with all three exacerbations in the 60 mg arm requiring hospitalization (mean 23.3 days) — a finding the authors flag in the context of TNIK inhibition’s potential immunomodulatory effects and the known link between AE-IPF and dismal survival (median 2.2 months post-onset).
• Sample size and duration limitations. 18 patients per arm, 12 weeks, China-only, 16 of 71 (22.5%) discontinuations across all arms. The paper explicitly states “a sample size calculation based on statistical power considerations was not performed” for FVC and the site-vs-central spirometry mix (29.6% of patients on site-sourced equipment, 70.4% on centrally-sourced SpiroSphere devices with central overread) is a real data-quality caveat that the paper does not fully resolve.

Rentosertib is, on the published record, the first drug with both its biological target and its molecular structure generated by AI to produce positive mid-stage human efficacy data. The qualifier “with both” is important: DSP-1181 (Exscientia / Sumitomo Dainippon Pharma, OCD) was the first AI-designed drug to enter clinical trials — Phase 1 in January 2020 — and was abandoned in 2022 after failing to meet Phase 1 study criteria. AU-007 (Biolojic Design, anti-IL-2 for solid tumors) is the first AI-designed antibody in Phase 2, confirmed in June 2026. No AI-designed drug of any kind has yet been FDA-approved. Rentosertib’s lead is real, but it is also a single positive read in a 71-patient, 12-week, single-region trial with a real safety signal in the combination arm.

The Phase 3 (NCT07687459) was announced on July 7, 2026 — a prospective, randomized, double-blind, placebo-controlled, parallel-group study of 320 IPF patients across 47 centers in China, with rentosertib administered orally for 52 weeks. The primary endpoint is the annual rate of FVC decline over 52 weeks — a slowing-decline endpoint, not an absolute-improvement endpoint. The trial is led by Professor Zuojun Xu of Peking Union Medical College Hospital as Leading PI, with Academician Nanshan Zhong of the Chinese Academy of Engineering and Professor Chang Chen of Shanghai Pulmonary Hospital as Co-Leading PIs. Expected completion is October 2029. The Phase 3 launch commentary from Insilico was led by Feng Ren, Ph.D., Co-CEO and Chief Scientific Officer of Insilico Medicine, who framed rentosertib as the output of a “biology-first, aging-informed AI workflow that connected TNIK to fibrotic and inflammatory disease mechanisms, and then used generative chemistry to create a drug candidate with the properties required for clinical development.” An inhaled formulation of rentosertib (delivered as a solution to the lung) received CDE IND clearance on April 28, 2026 — Insilico calls it the world’s first AI-driven candidate to enter a direct-to-lung clinical study. The inhalation program is the 13th Insilico PCC to receive IND clearance, and the 80-subject Phase 1 (SAD + MAD + IPF cohort) is now enrolling.

The Pipeline Beyond Rentosertib

As of June 30, 2026, Insilico’s pipeline comprises 31 nominated preclinical candidates, 13 with IND clearance, and 10 programs in clinical development (4 advanced independently + 6 progressed through collaborative clinical studies). The non-rentosertib clinical assets:

• Garutadustat (ISM5411) — a gut-restricted PHD inhibitor for inflammatory bowel disease. Granted the USAN generic name “garutadustat” in late 2025. Phase 2a BETHESDA (NCT07265570) dosed its first patient on January 11, 2026, with 80 estimated enrollment across approximately 80 participants with active ulcerative colitis. Endoscopy + histopathology + biomarker endpoints. Pre-clinical development published in Nature Biotechnology in December 2024.
• ISM6331 — a pan-TEAD inhibitor for mesothelioma and other solid tumors. Phase 1 (global multi-center, China + U.S.) first patient dosed January 22, 2025. Per Insilico, the molecule has a novel scaffold and “superior efficacy and safety” preclinically.
• ISM3412 — a MAT2A inhibitor for solid tumors (MTAP-deleted cancer). FDA IND cleared in 2024. Phase 1 first-patient dosing June 2025.
• ISM5939 — an oral ENPP1 inhibitor for solid tumors. Published in Nature Communications in 2025 as “Oral ENPP1 inhibitor designed using generative AI as next-generation cancer immunotherapy.” First-in-human Phase 1a/b (NCT06724042) start estimated 2026-06-30.
• ISM8969 — an NLRP3 inflammasome inhibitor co-developed with Hygtia Therapeutics, with dual IND approvals from China CDE and the U.S. FDA. First-in-human dosing in Phase 1 completed.
• ISM0676 — a GIPR antagonist. Per Insilico’s H1 2026 profit alert, demonstrated up to 31.3% weight loss in preclinical models.
• ISM6166 — a pan-KRAS inhibitor.
• ISM6200 — a selective NR3C1 antagonist (potentially best-in-class).
• ISM5059 — an NLRP3 inhibitor for systemic inflammatory diseases.
• ISM0387 — a PRMT5 inhibitor developed by Insilico’s UAE team, the first AI-driven innovative drug discovery milestone in the Gulf region.

Out-licensed assets:

• ISM3091 (XL309) — a USP1 inhibitor out-licensed to Exelixis on September 12, 2023, for an $80M upfront and up to $955M in development, commercial, and sales milestone payments plus tiered royalties. FDA IND cleared April 17, 2023. Exelixis is developing it as XL309.
• ISM5043 (MEN2312) — a KAT6 inhibitor out-licensed to Menarini Group (via Stemline Therapeutics) in December 2023. $12M upfront, up to $506M total milestones plus royalties. Indication: ER+/HER2- breast cancer.
• Undisclosed Menarini asset — December 2024, $20M upfront, up to $550M+ in milestones.
• ISM4808 — a PHD inhibitor out-licensed to TaiGen Biotechnology in late 2025 for anemia in chronic kidney disease. Total deal value in the “double-digit million” range.
• ISM8207 (QPCTL program) — co-developed 50-50 with Fosun Pharma; NMPA Phase 1 IND approved July 2023, Phase 1 trial initiated April 2024.

The Partnership Wave — 2025 to 2026

Insilico’s 2025-2026 partnership velocity is the strongest external validation of any pure-play AI-bio company in operation today. Headline numbers (deal value is total potential including milestones, not guaranteed cash):

• Sanofi (Nov 8, 2022) — multi-target, up to 6 targets. $21.5M upfront / target nomination fees + up to $1.2B total + tiered royalties. The deal is the longest-running of the cohort and is still active.
• Fosun Pharma (Nov 2021) — multi-program + QPCTL co-development. 50-50 R&D cost share on QPCTL. Multi-year collaboration.
• Exelixis (Sept 12, 2023) — ISM3091 USP1 global license. $80M upfront + up to $955M total. (Status noted above.)
• Menarini / Stemline (Dec 2023) — ISM5043 KAT6 global license. $12M upfront + up to $506M total.
• Menarini (Dec 2024) — undisclosed asset. $20M upfront + up to $550M+ total.
• Servier (Jan 4, 2026) — multi-year oncology R&D collaboration. $32M near-term + up to $888M total + cost-share.
• Eli Lilly (Mar 29, 2026) — global R&D collaboration, multiple therapeutic areas, exclusive worldwide license for oral preclinical programs. $115M upfront + up to $2.75B total + tiered royalties.
• SK Biopharmaceuticals (Jun 22, 2026) — CNS / neuroimmune disorders (neuroinflammatory, neurodegenerative, rare neurological). $18M near-term + up to $2.5B total + single-digit royalties.
• Takeda (Jul 1, 2026) — strategic collaboration, multiple therapeutic areas. $60M near-term + up to $600M total + tiered royalties.

Total headline deal value across the four 2026 deals (Servier + Lilly + SK Bio + Takeda): ~$6.74B in potential milestones, ~$225M in near-term cash. Add Sanofi’s $1.2B and the cumulative partner headlines (excluding the Exelixis and Menarini out-licenses) sit north of $7.9B in 12 months. Heavily backloaded on milestones — the cash actually collected in 2026 from these four deals is more like $225M, with $6.5B+ in milestones contingent on R&D, regulatory, and commercial readouts. The full Insilico prospectus discloses that at the IPO date, three out-licensing deals (Exelixis + Menarini x2) had a maximum total contract value of up to $2.1B including $110M in upfront payments and $1.9B in milestones, plus the multi-target collaborations with Sanofi, Lilly, and Fosun. The 2026 wave is a step-change on top of the 2022-2024 baseline.

The listed prospectus discloses that Insilico collaborates with 13 of the top 20 largest global pharmaceutical companies in terms of reported 2024 sales, up from “10 of the top 20” in pre-IPO marketing. The partnerships span software licensing, target ID, generative chemistry, and full asset out-licensing.

IPO and Financial Profile

Insilico listed on the Main Board of the Hong Kong Stock Exchange on December 30, 2025 under stock code 3696.HK, raising HKD 2.277 billion (approximately US$293M at the IPO offer price of HK$24.05 per share) — 2025’s largest Hong Kong biotech IPO and the first AI-driven biotech company to go public on the Main Board under Chapter 8.05 of the HKEX listing rules. 94,690,500 shares were offered globally, with a 10% Hong Kong public tranche oversubscribed by approximately 1,427.37 times (subscription funds of over HKD 328.349 billion) and a 90% international tranche oversubscribed by 26.27 times. 15 cornerstone investors included Eli Lilly, Tencent, Temasek, Schroders, UBS Asset Management, Oaktree, E Fund, and Taikang Life Insurance. Joint sponsors were Morgan Stanley, CICC, and GF Securities.

On July 8, 2026, Insilico released a positive profit alert for the first half of 2026, projecting revenue of approximately US$102.5M to US$106.5M (year-on-year growth of ~272.7% to 287.3%), net profit of approximately US$33.5M to US$39.5M, and adjusted non-IFRS net profit of approximately US$45.5M to US$51.5M. The financial alert is forward-looking and unaudited (the audited interim is expected in August 2026), so the H1 2026 swing from the prospectus-tracked track record (US$5.7M operating loss on US$27.5M revenue in H1 2025) to net profitability on US$100M+ revenue is real, large, and provisional. H1 2025 → H1 2026 was the half in which Insilico stopped running as a clinical-stage R&D shop and started operating as a profitable AI-bio commercial entity. The new revenue base is heavily partnership-driven (the 2026 deal wave), not software-subscription-driven, and the underlying $6.7B in headline deal value remains mostly backloaded.

The IPO use of proceeds: approximately 48% to fund clinical R&D for current clinical-stage programs, 20% for early-stage discovery, 15% for new generative AI model development, 12% for automated lab expansion, and 5% for working capital and other purposes.

Robotics and the Lab Stack

Insilico’s physical infrastructure is anchored by the Life Star robotics labs in Suzhou. Life Star 1 launched December 29, 2022 at Suzhou BioBAY as the first fully automated 6th-generation biology lab run by an AI-driven drug-discovery company, with six fully automated modules: sample management and QC, compound management, automated cell culture, high-throughput screening, high-content cell imaging, and next-generation sequencing. Life Star 2 launched in September 2025 with enhanced cross-island equipment coordination and multi-workflow parallel processing. The lab’s purpose is dual: (a) generate high-quality, proprietary experimental data to feed back into the Pharma.AI models for continuous retraining, and (b) accelerate wet-lab validation of AI-generated hypotheses. The company’s LabClaw agentic operating system (announced May 6, 2026) sits on top of Life Star 2 with 5 collaborative AI agents (Experiment Coordinator, Orchestration Expert, Science Analyst, QC Inspector, Data Specialist) and 28 specialized skill modules, governed by a Human-in-the-Loop approval mechanism for critical junctures. PandaClaw, announced in the H1 2026 profit alert, is the parallel agentic system for biological and bioinformatics engineering workflows.

The Critical Read

A few things the marketing-grade writeups in circulation are getting wrong or over-stating:

• “First AI-discovered drug” framing. DSP-1181 (Exscientia / Sumitomo Dainippon Pharma, OCD) was the first AI-designed drug to enter clinical trials (Phase 1 in January 2020) and was abandoned in 2022. Rentosertib is the first drug with both its biological target and its molecular structure generated by AI to produce positive mid-stage human efficacy data — a different claim.
• “Phase 2a primary endpoint was FVC.” It was not. The primary endpoint was TEAE rate (a safety endpoint). FVC was a secondary endpoint, the trial was not powered for FVC, the placebo 95% CI crossed zero, and 16 of 71 patients (22.5%) discontinued across all arms. The Nature Medicine paper itself states “a sample size calculation based on statistical power considerations was not performed.”
• “Phase 2a failed primary endpoint in 2024.” Wrong. The trial met its primary safety endpoint; the November 12, 2024 topline was positive; the June 3, 2025 Nature Medicine publication reported positive results. The “failed” framing in some prior writeups is a misreading of the November 2024 topline release.
• “Lilly deal worth $2.5B in 2025” is wrong. The 2026-03-29 Lilly deal is worth up to $2.75B. The 2025-11 announcement was an expansion of an existing collaboration at “more than $100M” terms.
• “SK Bio $2.5B in 2025” is wrong. The $2.5B SK Biopharmaceuticals deal is dated June 22, 2026 at BIO 2026 in San Diego.
• “Insilico’s lab is in Boston.” Life Star 1 and 2 are in Suzhou BioBAY. The Cambridge, MA site is corporate, not the robotics lab.
• “13 of top 20 pharma are paying SaaS customers.” The 13-of-top-20 figure (post-IPO prospectus) refers to collaborations, not full SaaS contracts. Software solutions were 2.8%-6.6% of revenue across the prospectus track record period; the $4.74B headline deal value is from co-discovery and out-licensing agreements, not from software-license revenue.
• “Generative AI designed a drug end-to-end” is closer to accurate, but the wet-lab validation, medicinal-chemistry optimization, manufacturing, and clinical development are still done by humans, CROs, and CDMOs. AI proposes; humans and biology decide. Of the 78 molecules synthesized in the original rentosertib discovery, every one was made in a wet lab and characterized in in-vitro and in-vivo assays before a single patient was dosed.
• “Phase 3 will confirm Phase 2a.” The Phase 3 is a different design — 52-week, 320-patient, China-only, primary endpoint = rate of FVC decline (not absolute improvement). It is not a guaranteed win. The drug-drug interaction with nintedanib will need separate study.
• “H1 2026 net profit $33-40M” is a forward-looking profit alert, not an audited result. Audited interim is expected August 2026. Provisional.
• “No AI-designed drug has yet been approved.” True. No AI-designed drug of any kind — rentosertib, DSP-1181, AU-007, or any other — has been approved by the FDA, EMA, China NMPA, or any major regulatory body as of mid-2026.

What to Watch Over the Next 6 Months

• Late Q3 2026 — Insilico Phase 3 52-week enrollment updates. China-only 320-patient trial; enrollment pace is the gating signal. The protocol’s primary endpoint (rate of FVC decline) is materially different from the Phase 2a secondary FVC signal.
• Q4 2026 — Rentosertib inhaled Phase 1 PK readout. The April 28, 2026 CDE-cleared Phase 1 for the direct-to-lung formulation is the 13th Insilico PCC to receive IND clearance. Topline safety and PK data will set the clinical-development stage for the 52-week inhaled efficacy design.
• Q1 2027 — H1 2026 audited interim (expected August 2026, restated). The forward-looking H1 2026 profit alert projects net profit of US$33-40M on US$102-107M revenue. Audited interim is the canonical data point.
• H1 2027 — Servier program candidate. The $888M oncology R&D collaboration with Servier has been running since January 2026. The first Insilico-discovered candidate from the Servier collaboration advancing to PCC nomination is the next partnership value catalyst.
• Q1-Q2 2027 — Lilly program updates. The $2.75B Lilly collaboration has been running since March 2026. The first candidate from the Lilly collaboration is the next partnership value catalyst.
• Throughout 2026-2027 — competitive AI-bio Phase 3 readouts. Generate Biomedicines’ SOLAIRIA-1 (Phase 3, AI-designed antibody) reads out late 2027/early 2028. If rentosertib hits Phase 3 in 2028-2029, it would be the first AI-discovered and AI-designed small molecule to clear a pivotal trial. The bar matters: rentosertib would be the validation case for the entire AI-bio thesis, not just for Insilico.

The Bottom Line

Insilico Medicine has the most clinically validated end-to-end generative AI drug-discovery platform in the world right now. The Pharma.AI suite is the only AI-bio platform with a peer-reviewed, prospective, mid-stage human efficacy validation in a top-tier journal. The Phase 2a GENESIS-IPF results are real, well-controlled, and the first of their kind — but they are small, short, exploratory on FVC, and carry a real liver-toxicity signal in nintedanib combination. The Phase 3 protocol is a different bar. The HKEX 3696 listing in December 2025 turned a 10-year R&D burn into a public-market capital structure with $293M in new cash. The 2026 deal wave — Servier, Lilly, SK Biopharmaceuticals, Takeda — added $6.7B in headline deal value and pushed Insilico into a position where the company is profitable on a forward-looking H1 2026 basis for the first time.

The combination of (a) AI capability, (b) proprietary wet-lab data loops through Life Star 2 + LabClaw, (c) at least one clinical-stage asset with positive mid-stage human data, and (d) commercial pharma partnerships with 13 of the top 20 global pharma is the canonical four-bucket recipe for AI-bio survival. Insilico has all four. The 2026-2027 question is not whether the AI works — it works, and the Phase 2a data proves it. The question is whether the Phase 3 trial design, the liver-toxicity profile in combination therapy, and the regulatory pathway can take rentosertib to a first AI-discovered-and-designed drug approval.

If rentosertib reads out positive in 2028 or 2029, Insilico becomes the validation case for the entire AI-bio thesis, and the $6.7B in 2026 partnership value will look cheap in hindsight. If the Phase 3 misses or generates a smaller FVC signal than the Phase 2a, the rentosertib narrative becomes a cautionary tale about sample size, combination toxicity, and the gap between “AI discovered the molecule” and “AI discovered the drug.” Both outcomes are realistic. The next 24-30 months of data will tell.

---

Sources: Ren et al., Nature Biotechnology (online March 8, 2024, print 43:63-75, 2025, DOI 10.1038/s41587-024-02143-0, PMID 38459338); Xu, Ren, Wang et al., Nature Medicine 31:2602-2610 (June 3, 2025, DOI 10.1038/s41591-025-03743-2, PMID 40461817); Ivanenkov et al., J. Chem. Inf. Model. 63:695-701 (February 2, 2023, DOI 10.1021/acs.jcim.2c01191, PMID 36728505); Kamya et al., J. Chem. Inf. Model. (February 25, 2024, DOI 10.1021/acs.jcim.3c01619); Aliper et al., Clin. Pharmacol. Ther. 114:972-980 (2023, DOI 10.1002/cpt.3008, PMID 37483175); Livne et al., Chem. Sci. 15:8380-8389 (2024, DOI 10.1039/d4sc00966e, PMID 38846388); Zhavoronkov et al., Nature Biotechnology (2019) [GENTRL / DDR1]; Ozerov et al., Nature Communications 7:13427 (2016) [iPANDA]; Insilico Medicine HKEX prospectus (December 18, 2025); Insilico Medicine press releases: HKEX listing (December 30, 2025), Servier collaboration (January 4, 2026), BETHESDA first patient dosed (January 11, 2026), MMAI Gym launch (January 2026), DORA Community Edition on GitHub (January 8, 2026), Lilly global R&D collaboration (March 29, 2026), rentosertib inhaled CDE IND clearance (April 28, 2026), LabClaw announcement (May 6, 2026), SK Biopharmaceuticals collaboration (June 22, 2026), Nach01 launch on AWS Marketplace (June 10, 2025), Takeda collaboration (July 1, 2026), Phase 3 initiation (July 7, 2026), H1 2026 positive profit alert (July 8, 2026); Exelixis IR press release ISM3091 license (September 12, 2023); Menarini Group press release KAT6 license (January 2024) and second oncology license (December 2024); Sanofi–Insilico $1.2B collaboration (Qiming Venture Partners / FierceBiotech, November 8, 2022); Reuters and STAT coverage of the Eli Lilly deal (March 29-30, 2026); Forbes coverage of HKEX listing (December 30, 2025); ClinicalTrials.gov records for NCT05938920 (Phase 2a GENESIS-IPF), NCT07687459 (Phase 3), NCT05154240 (Phase 1), NCT07265570 (BETHESDA), NCT06724042 (ISM5939 Phase 1); Hong Kong Investment Corporation, Morgan Stanley, CICC, and GF Securities as joint IPO sponsors; weekly industry research sweeps of biotech partnership and clinical-trial data.