Ajay Gopal, PhD

(nearly) Everything you need to know for antibody developability & property prediction, a guide At Tamarind Bio, we gathered best practices + the most effective (mostly open) in silico tools across all key risks: Aggregation propensity: why it kills programs; sequence + structure tactics (Aggrescan, TANGO/Waltz, Aggrescan3D; HIC-ML surrogates). Solubility: sequence and structure predictors to rank and design mutations (NetSolP, Protein-Sol, CamSol, SOLpro/SOLart; DeepSol/PON-Sol). Viscosity & self-interaction: charge/hydrophobic patch analysis (SCM, DeepSP), B22/kD surrogates; practical cut-offs for high-conc mAbs. Stability & chemical liabilities: ΔΔG and hotspot scanning (ThermoMPNN; TEMPRO for VHH); deamidation/isoAsp, oxidation, glyco motifs, free Cys (abYsis, MusiteDeep). Immunogenicity: HLA-II centric risk scoring and deimmunization workflow (TLimmuno2, NetMHCIIpan, IEDB; B-cell: DiscoTope, ElliPro, ABadapt/EpiPred). pI & charge profile: fast screens + patch asymmetry (IPC/ProtParam; TAP, SCM). Manufacturability & expression: predicting titers and downstream pain via solubility/patch proxies; chromatography behavior cues (HIC/IEC). We also detail best practices for multi-stage filtering, evaluating multiple parameters e.g. solubility, stability, aggregation, distributions vs. clinical stage mAbs, along with how to benchmark and validate these tools. Read the full guide: https://lnkd.in/gS7rx4JK ——— Tamarind Bio is leading the collection of bioinformatics and computational molecular design tools like AlphaFold, RFdiffusion, (Ig)BLAST, ProteinMPNN accessible via our scientist-friendly web interface, AI Copilot and programmatic API.

228

7 Comments

Introducing Boltz-2, an open-source model from MIT and Recursion that predicts both protein structures and binding affinities. It reaches free-energy-perturbation (FEP) accuracy at ~1,000× the speed. Use it on Tamarind Bio today: https://lnkd.in/gqnAG8Tz Binding affinity determines whether a drug is potent enough; FEP is precise but slow, while docking is fast but noisy. Deep-learning alternatives have lagged, until now. Performance: • Matches OpenFE on the standard FEP+ benchmark while running 1,000× faster. • Tops the CASP16 affinity challenge (140 complexes). • In MF-PCBA hit-discovery screens, doubles average precision over ML and docking. Coupled with SynFlowNet, Boltz-2 powered a prospective TYK2 screen; ABFE predicted binding for all top-10 generated compounds. For structure prediction, Boltz-2 Matches or exceeds Boltz-1, with major improvements on DNA–protein, RNA, and antibody–antigen complexes. Specify contact constraints, templates, or experimental methods directly in the model. Architecture upgrades: Builds on Boltz-1 with a new affinity head, finer controllability, GPU optimizations, and extensive synthetic + MD training data. Boltz-2 brings FEP-level accuracy to deep learning, at a fraction of the compute cost, and slots cleanly into generative screening pipelines. Excited to see how Boltz-2 will be used for faster small molecule discovery! Code & model: https://lnkd.in/ezDA9pji Paper: https://bit.ly/boltz2-pdf

722

11 Comments

Exciting news today from AWS partner Latent Labs… Latent-X2, a frontier AI model, can design drug-like biologics without iteration. The generated designs display drug-like properties including low ex vivo immunogenicity, significantly shortening the path from hit to clinical candidate. Congrats, Simon Kohl and team! https://lnkd.in/gpMxN-Xh

131

1 Comment

🆕 #NoPriorsPod! AI and atomic-level structure prediction are accelerating drug discovery — turning a slow, empirical search process into something closer to molecular engineering. Joshua Meier and Jack Dent co-founders Chai Discovery chat their new bio foundation model, Chai-2. It enables zero-shot antibody discovery in a single 24-well plate — with a ~20% hit rate, compressing years of lab work into weeks. They talk what's next for Chai, what's still challenging, and why they're 𝐛𝐮𝐥𝐥𝐢𝐬𝐡 𝐨𝐧 𝐛𝐢𝐨𝐭𝐞𝐜𝐡

88

5 Comments

A year ago, we launched Inductive Bio with the belief that AI, while not a silver bullet, can—and should—make drug discovery dramatically faster, cheaper, and more successful. Today, I'm excited to share that we’ve raised a $25M Series A to accelerate that mission. At Inductive, we know that models are a piece of the puzzle, but not everything. To make the promise of AI a reality,  we’re building infrastructure the whole industry can benefit from: 🔹A pre-competitive data consortium with diverse data from across chemical space 🔹State-of-the-art AI models to predict ADMET properties before synthesis, guiding better drug design 🔹Compass, an intuitive software platform already powering millions of molecule designs and dozens of real-world programs The path from idea to therapy is still far too slow and uncertain. Our goal is to help scientists make better decisions earlier — and unlock therapies that otherwise might never reach patients. I’m grateful to our team, our partners, and the investors joining us: Obvious Ventures, a16z Bio + Health, Lux Capital, S32, Character Capital, Amino Collective, and an inspiring group of angels. We’re just getting started. 👉 https://lnkd.in/ev7h9QTU #DrugDiscovery #AI #Biotech #Startups #LifeSciences #InductiveBio

447

104 Comments

AI Agents will become a key tool for scientific discovery and a defining biotech trend for 2025. They're gaining real momentum doing complex scientific tasks—and now autonomously generating biological hypotheses! Check out Genentech’s new preprint on SpatialAgent, an LLM-based AI agent built for full-stack spatial biology analysis (link in comments). SpatialAgent combines memory, planning, and action modules, navigating datasets, tools, and biological systems without predefined workflows. The preprint is hefty at 110 pages—but a few (Anara-derived 😆) highlights stand out: - Agent outperformed human experts in gene panel design and spatial annotation across multiple tissues. - Integrates toolchains dynamically, including scRNA-seq references, anatomical images, and interaction databases. - Generates interpretable reasoning, enabling real-time hypothesis generation and iterative discovery. Where Tech Meets Bio did a great landscape on this space highlighting other agentic systems such as Stanford’s BioDiscoveryAgent (Yusuf Roohani) and Harvard’s TxAgent (Shanghua Gao) plus others. What are your favorite use cases for AI agents in Bio? Let's chat if you're working on new ones 👋 ! #AI #Biotech #AgenticAI #HealthcareInnovation Image source: “SpatialAgent: An Autonomous AI Agent for Spatial Biology” (link in the comments)

120

23 Comments

Had fun over the last few days writing this blog about batch effects in biological data and how they impede AI advancement in biology. I was so excited about the topic that I transformed it into an AI podcast using a combination of #kling and #google notebook LLM. My exploration reveals a fascinating contrast: while genomics and protein structure prediction (like AlphaFold) have seen remarkable AI success due to their standardized or evolution-based data, most other biological measurements remain plagued by batch effects. This explains why we see GenomeGPT and protein models proliferating while other biological domains struggle—yet these successful applications barely scratch the surface of what's possible at the bio+AI intersection. The blog examines two potential solutions for overcoming these challenges: radical transparency in experimental documentation or manufacturing-grade standardization of data generation. Though these approaches face significant adoption barriers in a field still struggling with its reproducibility crisis, they represent our best path toward unlocking the true potential of biological AI. More than thought on the content itself, I'd love your thoughts on which format did you find more engaging—reading the blog or listening/viewing the AI-generated podcast? Link to full blog: https://lnkd.in/gZ7k9jY3 #AIinBiology #Bioinformatics #MachineLearning #BatchEffects #HealthTech #DataScience

45

6 Comments

Love this diagram from Brendan Keeler's substack on AI in Healthcare. Apps like Citizen will be focused on helping the sickest patients and their families navigate health, but it requires all players to get on board. It's the sickest patients who really need better access to data and tools that help them manage complex and difficult conditions. https://lnkd.in/gu9n2cyU

68

5 Comments

Startups like Triomics and Snorkel AI are turning AI research into real-world impact, from helping match cancer patients with clinical trials to building smarter data labeling systems that accelerate enterprise AI adoption.   These stories reflect the power of the Bay Area ecosystem and the role Microsoft plays in enabling breakthrough innovation through platforms like Azure.   Read more on the Microsoft Bay Area blog: https://lnkd.in/eUawtCvp Annie Pearl Microsoft for Startups M12, Microsoft's Venture Fund #Healthcare #DataAndAI #Azure #MicrosoftBayArea #AIInnovation

31

1 Comment

Everyone wants to build the AlphaFold of hit discovery, and Target 2035 is a bold attempt. Mapping binders to every human protein, with open data and tools like AIRCHECK and CACHE, is a commendable step. But let’s not confuse binders with drugs. DEL and AS-MS, the primary methods behind Target 2035, weren’t designed for training reliable ML models for drug discovery or functional biology; they detect binders, but not binding modes or biological function. In my view, what we need instead are functionally validated, target-specific datasets. That means: ▪️ high-quality, structure-prepared target proteins, ▪️accurate docking and binding mode predictions, ▪️and robust pocket selection strategies. With this foundation, we can identify hits with MoA, not just statistical binders. In some cases, a binder is enough (e.g., degraders, disruptors), but for most drug discovery, mechanism matters. I support the open-science ambition of Target 2035, but its utility for model training depends on enriching the biological signal. Models won’t learn biology from assay artifacts. Paper: https://lnkd.in/eqhwMPFQ #opensource #drugdiscovery #ML #openscience

47

1 Comment

"If I can monitor my Amazon package or Domino's pizza, I should be able to monitor my $4.2M gene therapy." But how does that work? Rohin Krishnan Iyer ,. provides an overview of how single use technologies are being used in the supply chain & logistics to provide real-time data on therapy batches - regardless of where they are in transit. This clip was taken from a full interview on supply chain logistics at Phacilitate's ATW. You can view the full episode here: https://lnkd.in/g845Wm6R What are your thoughts?

116

10 Comments

tl;dr - I built an OpenClaw plugin for ŌURA data, and it's bonkers. One of the biggest challenges and opportunities with LLMs is to ensure that they have rich context for personalized responses. This is why OpenClaw - which has gone from zero to hero in just a few weeks - has tremendous potential (also huge risks - more on that below). If you don't know what I'm talking about, OpenClaw (formerly Clawdbot/Moltbot) is an open source project that runs on your computer and allows integration with a rapidly-growing set of tools/plugins/skills that the LLM of your choice (e.g., Anthropic, Google, xAI, OpenAI, or local models) can interact with. Imagine a personal assistant that can check-in to your flights on your behalf, install software with a simple request, or provide a summary of your day every morning. And this is all done through the messaging service of your choice: iMessage, WhatsApp, Signal, Telegram, etc. This is basically Iron Man's Jarvis, available today. I was curious how well OpenClaw would work with the high-quality, continuous data provided by Oura Ring, so I built a plugin (open source, here: https://lnkd.in/eXDZhZRs) that makes it easy for OpenClaw to query your data when it decides it would be useful. That last bit is key - modern LLMs can decide which tools would be best to answer a particular question. Disclaimer: this is a personal project and not endorsed by Oura. Use at your own risk. :) The plugin requires a few steps to set up, but it walks you through them. You can decide if you want a morning and evening summary of our Oura sent to your preferred messaging platform, and/or you can simple use it to ask questions. There have already been a few magical moments. The first was when OpenClaw proactively used my calendar data (via another plugin) to nudge me to get to sleep earlier: "💡 Sleep balance has been low for a couple days. Consider an earlier bedtime tonight — you've got swimming at 6:30 AM tomorrow." The second was when it proactively asked if it should track my sleep for a week and then follow-up afterwards (see screenshot) - real accountability! WARNING: if you don't know what a command line is, you probably shouldn't be using OpenClaw. There are real risks, including prompt injection, exfiltration of data, malicious plugins - you name it. I've set this up on a separate personal computer with local loopback and with a new/dedicated Apple Account and limited access to my actual data (including no access to my work data, and of course no connections to any AI social networks **ahem, Moltbook**!). But I learn by doing, and this is the new frontier. If you know what you're doing and want to try it out, I'd love your feedback!

154

16 Comments

The EU's Executive VP for Tech Sovereignty recently shared the AI Continent Action Plan, which focuses heavily on "AI Factories" to provide computing power for European companies. As a top 5 biotech & AI founder in Europe, and having been responsible for parts of Google's ML infrastructure, I find this being front in center in the European AI strategy concerning: It's old wine in new bottles. These "AI Factories" are essentially EuroHPC supercomputers being rebranded for AI. But there's a fundamental mismatch: • They're built for physics simulations, not AI workloads • They're inflexible compared to modern cloud infrastructure and lack a bunch of other relevant services to build products • Depending on the age, but in most cases they will require significant rework of models and infrastructure software to make them even work. Which is not something you should be spending time on as a startup. Most European AI scale-ups seem to avoid these systems entirely (if you do, please comment), because existing cloud solutions work pretty well, and they're actually built for purpose. If we want to compete globally in AI, lets focus on asking founders of scale ups what they need to succeed. If we worry about EU Cloud Infrastructure sovereignty, we should try to find ways to be able to de-couple the sovereignty risks current cloud providers pose. For example, we could put European cloud provider infrastructure under separate legal entities with strict legal and technical controls to manage sovereignty risk. Thoughts?

176

24 Comments