Apheris

Our CEO and co-founder Robin Roehm will be in San Francisco for JPM Week 2026. 🎇 This comes as the AI Structural Biology (AISB) Network has reached an important milestone with its latest results. In this initiative, AbbVie, Astex, Bristol Myers Squibb, Johnson & Johnson, and Takeda fine-tuned OpenFold3 on several thousand proprietary structures, creating the largest and most diverse industrial, structural dataset ever used for co-folding model training. Apheris’ federated computing product connects the data and ensures it stays fully protected inside pharma's own IT environments. Check out the Science News coverage on the AISB Network (see link in comments) 👉 Data is the differentiator: Training on more diverse, proprietary data from industry partners clearly improves co-folding model performance. Far beyond what any single party can achieve. 🔎 Apheris builds life sciences data networks and the infrastructure that make collaborations like AISB possible. AISB is welcoming new partners to the network and is extending its Federated OpenFold3 Initiative. 🗨️ If you’re joining JPM Week and want to discuss co-folding, federated training, or upcoming network initiatives, reach out. #StructuralBiology #CoFolding #FederatedLearning #JPM

Welcome to the Apheris team, Rosemary Huckvale! Rosemary joins us as a medicinal chemist and will work on integrating our co-folding capabilities into the workflows of computational and medicinal chemists. She studied chemistry at the University of Oxford and completed her PhD in synthetic chemistry at University College London before training as a medicinal chemist. She has held roles at The Institute of Cancer Research, Healx, and CHARM Therapeutics, contributing to early discovery and compound design. Her background gives her a practical understanding of how chemists use structural information to generate and prioritize hypotheses, an important perspective for shaping our product. 💬 Why she joined Apheris “I’ve seen in my previous work how training AI models on diverse, proprietary data meaningfully improves structural predictions, which makes the idea of federated networks compelling. When predictions are stronger, medicinal chemists can make decisions with more confidence and test hypotheses before moving ideas into the lab. I’m excited to help build this capability with the Apheris team.” We’re pleased to welcome you on board and look forward to advancing the next chapter of structure-based drug discovery together. #MedicinalChemistry #StructureBasedDrugDiscovery #CoFolding #Hiring

Great paper from Lucy Vost, Yael Ben-Gigi Ziv, and Charlotte Deane on what it takes to design reliable structure-based small molecules. Three main conclusions: 1. You need accurate protein structures first Small geometric errors propagate directly into binding and stability ranking. If the structure is off, everything built on top of it becomes unreliable. 2. Models fail where structural data is thin Methods look strong on well-represented folds but break once the chemistry or binding mode drifts from what the model has seen. 3. Benchmarks must match real drug discovery systems Most evaluation sets are too narrow. The authors argue for datasets that reflect the actual targets and binding situations medicinal chemists deal with. 🔗 Link to the paper: https://lnkd.in/ePG2sTX5 This mirrors exactly what we see across the field and what we are addressing with the AI Structural Biology (AISB) Network. In the Federated OpenFold3 Initiative, Abbie, Astex, BMS, Johnson & Johnson, and Takeda contributed several thousand experimentally determined protein–small molecule structures each. These datasets reflect industrial discovery chemistry, not just public folds, and OpenFold3 is trained on this combined structural dataset through our federated computing product. All data staying local and all IP protected. #structuralbiology #drugdiscovery #computationalbiology

We’re introducing ApherisFold Lite, a browser-based version of ApherisFold that you can use right away to run OpenFold3 without any setup. 🔎 What ApherisFold Lite offers ApherisFold Lite gives you a simple way to experience the ApherisFold interface and run co-folding predictions with OpenFold3. It is intended for early exploration to help you: • Understand workflows • Review outputs • Assess whether the product fits your internal research processes before setting up a full ApherisFold deployment 🎇 What the full ApherisFold product provides The full ApherisFold product delivers a complete co-folding capability. It runs entirely within your environment and enables: • Secure local inference on in-house data • Systematic benchmarking using public or proprietary datasets • API-based integration into existing discovery workflows • Fine-tuning co-folding models on proprietary structures or via federated data networks These functions are essential for establishing where a model is reliable, where it fails, and how it can be adapted or improved. 👉 ApherisFold Lite offers a low-friction way to try the product; ApherisFold delivers the full capability. Explore ApherisFold Lite: https://lnkd.in/eSS6f5FJ #ProteinStructurePrediction #CoFoldingAI #StructuralBiology

We had our Apheris Office Days this week. In a remote-first company, these days carry a lot of weight. They set the tone for the months where we only see each other through screens and give us a chance to understand each other outside of structured calls. People’s humor, how they think, what they care about. New colleagues from pharma, techbio, and ML joined this quarter. They add to an already strong mix of experience in the team. Curiosity, thoughtful questions, and a willingness to step in and support each other were visible throughout the week! Qualities that have long shaped how we work together. The team ran sessions on topics central to our work, for example: - Calum Hand on Inference differences between ADMET and Co-folding and what's useful to consider - Mees Hendriks on structure prediction metrics - Ruda Filgueiras on fixing a Tensor Flow issue that arised in large-model federated training We also made time for the lighter parts: a Secret Santa book swap, a dinner on a boat, and the many small moments in between. Our latest people survey reflected very strong results in areas that matter a lot to us: trust, psychological safety, and the sense that people care about their work and collaborate. A good week, and now everyone is heading home tired, slightly overloaded, and hopefully a bit more connected. Also really grateful for the visits of our old colleagues Jan Stücke and Markus Bujotzek. Some moments captured in the photos below with: Daniel Griffiths José-Tomás (JT) Prieto Ian Hales Benedict W. J. Irwin 奔小康 Alex Jamieson-Binnie, Ph.D. Avelino Javer Christoph M. Evelyn Trautmann Nicolas Gautier Christopher Woodward Ariana L. Marie Roehm And last but not least huge shoutout to Ulrike Rahn for pulling the entire week together #DrugDiscoveryAI #FederatedLearning #RemoteCulture

We're excited to welcome two new colleagues to the Apheris team! Both will support us in advancing our co-folding application and building out our life sciences data networks by contributing their expertise in structural biology, cheminformatics, and model development. Here’s who just joined us: 🧬 Alex Jamieson-Binnie, Ph.D. – Lead Data Engineer – Structural Biology Alex joins us after three years at CHARM Therapeutics, where he applied his PhD in computational chemistry to creating structural data pipelines and developing tooling for using co-folding models in drug discovery projects. His work includes data preprocessing, cheminformatics & user interface design — areas that are essential for reliable model training across distributed datasets. 💬 “I’m very excited to join a team with strong engineering expertise, applying this to the challenging field of co-folding. I’m particularly interested in applying myself to the challenge of data preparation from diverse sources to help take these models to the next level.” 🧪 Mees Hendriks – Research Engineer – Drug Discovery Mees also joins us from CHARM Therapeutics. He contributed structure metrics to CHARM's DragonFold, helped design and implement the post-inference pipeline that applies cheminformatics methods to predicted structures, and built CHARM’s internal virtual chemical space platform together with medicinal and computational chemists. Mees holds an Msci in Natural Sciences from the University of Cambridge. 💬 “Apheris has delivered a state-of-the-art federated cofolding platform. I am excited to bring my expertise in developing methods and applications to bring transparent benchmarking and I am looking forward to develop the application for end-user accessibility." We’re very happy to welcome Alex and Mees to the team. #DrugDiscoveryAI #StructureBasedDiscovery #CoFolding #FederatedLearning #TeamGrowth

We couldn’t be happier to celebrate your journey with us over the past six months 🚀 From day one, you jumped straight into one of the most complex areas of our work and integrated into the team as if you’d always been part of it. Your federated diversity analysis, your paper, and the presentations you delivered at FLTA in Dubrovnik and at NVIDIA FLARE Days were such a great contribution! You're already missed at our Berlin office, so already looking forward to having you join our office days next week 🤗 And of course, we’re cheering you on as you wrap up your PhD, wishing you all the focus, motivation, and great breakthroughs for the final stretch! 🎓✨ Go Markus Bujotzek 👏 Thanks for being an impactful part of Apheris — once family, always family.💙

Apheris 3.9 is here, a big release built for our customers running federated AI and analytics in production across pharma, healthcare, and research networks where data can’t move, but insights must. Here’s what’s new (and why it matters): 💾 Reusing model results without retraining Federated experiments generate huge amounts of results: model weights, checkpoints, metrics — that are often lost after initial training runs. With the new Data Artifacts Store, everything can now be securely stored, reused, or fine-tuned across runs. ⚙️ Gateways that keep themselves up to date Our federated compute infrastructure now supports automatic upgrades using Flux controllers and Helm charts. That means less manual work, consistent versions across collaborators, and faster deployment of security and performance updates. 📈 Simpler models, cleaner pipelines We’ve added a Linear Regression model for interpretable analytics and refactored preprocessing into its own lightweight package, apheris-preprocessing, for easier integration and cleaner dependencies. 💡 Why this release matters Our customers are scaling federated AI across hospitals, research sites, and life sciences organizations and they need a product that’s stable, scalable, and easy to maintain. Apheris 3.9 is built for you, incorporating our joint learnings over the past months. 🙌 Huge thanks to our customers for continuously pushing us to make federated computing more practical, and to our engineering and product teams for bringing these features to life. 🔗 Read the full release blog: https://lnkd.in/exNWefJv #FederatedComputing #PrivacyPreservingAI #LifeSciencesData

Our colleague Benedict W. J. Irwin 奔小康 will be speaking at the UKQSAR Autumn Meeting in Cambridge (AstraZeneca Discovery Centre) about: 💡 Federated Learning for Protein Co-Folding Models Using Open-Source AlphaFold3 Derivatives. Ben will share an overview about the Federated OpenFold3 Initiative, launched by the AI Structural Biology (AISB) Network, where five leading pharma companies collaboratively fine-tuned OpenFold3 on their molecular structures using Apheris. If you’re at #UKQSAR2025, don’t miss Ben’s session on Nov 13th at 13:30 (Session 2) #FederatedLearning #QSAR #CoFoldingAI #DrugDiscovery

At yesterday's NVIDIA FLARE Q4 Webinar, Roche and Apheris presented two use cases: 👉 multi-site multiple sclerosis lesion segmentation and the federated OpenFold3 initiative. The Roche study has been published in Frontiers in Neurology, with Apheris powering the federated technology that made cross-hospital collaboration possible. 👉 “Federated learning for lesion segmentation in multiple sclerosis: a real-world multi-center feasibility study” by Hindawi et al.: 🔗 Read the paper: https://lnkd.in/enJZyNwE 🔗 Watch the webinar recording: https://lnkd.in/eAxqf6-c 👏 Thanks Frederik Buijs Eric Boernert Chester Chen and Holger R. Roth and of course to the Apheris team Nicolas Gautier and Robin Roehm for the insightful discussion and presentation #FederatedLearning #DrugDiscovery #MultipleSclerosis