Drug Discovery Methods Beyond Quantum Computing

Machine Learning in Preclinical Drug Discovery 🧬💊 Machine learning (ML) is increasingly integrated into preclinical drug discovery, offering promising advancements across hit identification, mechanism-of-action elucidation, and translational investigations. A recent paper in Nature Chemical Biology, "Machine Learning in Preclinical Drug Discovery", provides a thorough analysis of how ML is being utilized to enhance efficiency in early-stage drug development. 🔬 Key Insights from the Paper 1️⃣ Hit Identification & Virtual Screening Traditionally, high-throughput screening (HTS) has been the gold standard for identifying potential drug candidates. However, it is resource-intensive and slow. ML-based virtual screening, powered by deep learning models and molecular featurization techniques, is enabling rapid exploration of chemical libraries far beyond what traditional HTS can achieve. The paper highlights the impact of message-passing neural networks (MPNNs) and Deep Docking as effective methods for prioritizing hit compounds. 2️⃣ Mechanism-of-Action (MOA) Elucidation Understanding how a compound interacts with biological targets is critical for drug development. ML is now playing a pivotal role in MOA elucidation through: AlphaFold and RoseTTAFold: AI-driven protein structure prediction is accelerating target identification and binding site analysis. Generative models: Variational autoencoders (VAEs) and diffusion models are not only aiding in de novo drug design but also helping predict chemical interactions with biological systems. 3️⃣ Translational Investigations & ADMET Predictions Many promising compounds fail in later stages due to poor pharmacokinetics and toxicity profiles. ML is being leveraged to enhance ADMET predictions, improving the likelihood of clinical success. The paper discusses advancements in: Solubility and Lipophilicity Predictions: ML-driven models now outperform traditional log(P) estimations, increasing the reliability of early-stage compound selection. Toxicity Screening: AI-powered tools are improving predictions of hERG binding and organ toxicity, reducing late-stage failures. 🚀 The Future of AI in Drug Discovery While ML is proving to be a game-changer, challenges remain, including data quality, interpretability of AI models, and integration with experimental validation. The paper underscores the importance of open-source datasets, AI transparency, and active learning strategies to enhance model accuracy. 🔗 Read the full paper here: https://lnkd.in/gMtXHrHi AI is reshaping the landscape of drug discovery. As these technologies evolve, collaboration between computational scientists, biologists, and chemists will be critical to unlocking their full potential. #AI #MachineLearning #DrugDiscovery #Pharma #Biotech #ArtificialIntelligence #ComputationalBiology #NatureChemicalBiology

In this study just published (https://lnkd.in/eRRkhDgY), we asked the question: Can whole-brain imaging of c-Fos expression be used for predicting and classifying drugs?   We tested a panel of compounds: -      Psilocybin -      Ketamine -      5-MeO-DMT -      6-fluoro-DET (a non-hallucinogenic 5-HT2A agonist) -      MDMA -      Fluoxetine, acute administration -      Fluoxetine, chronic administration -      Saline vehicle   We used tissue clearing and light-sheet fluorescence microscopy to image c-Fos+ cells in the whole mouse brain (see the beautiful image!). This yielded a very large data set of 316 brain regions x 48 animals across the 8 drug conditions. We developed a computational analysis pipeline, incorporating the statistical Boruta procedure for region selection and ridge logistic regression for classification, while making sure there was no data leakage.   The classification was fairly accurate. If we ask the computer to predict the exact drug based on only the brain-wide c-Fos signals, it predicts with 67% accuracy, substantially above chance level (12.5%). Models that are designed to classify between specific pairs of drugs can do even better, predicting almost perfectly(>90%). We implemented Shapley additive explanation to identify the brain regions driving the drug classification. We found that the models work well with only few brain regions – for instance, prediction of psilocybin vs. 5-MeO-DMT relied on c-Fos signals in ~20 brain regions, including lateral habenula, and regions important for vision and arousal.   As a lab, we are keen to develop new methods for drug discovery, especially based on information that came from a strong neurobiological basis.   This was a project led by a pair of MD/PhD students, Farid Aboharb, MD PhD and Pasha Davoudian. The study would be impossible without the compounds from Alexander Sherwood at Usona Institute, and support from National Institute of Mental Health (NIMH), The National Institute on Drug Abuse (NIDA), and One Mind.

AI-powered virtual screening that scores 10 trillion protein-ligand pairs in a single day Of ~20,000 human protein-coding genes, only about 10% have been successfully targeted by FDA-approved drugs or have documented small-molecule binders. The bottleneck isn't biology—it's computational scale. Traditional molecular docking takes seconds to minutes per protein-ligand pair, making genome-wide screening essentially impossible with current resources. Yinjun Jia and coauthors tackle this head-on with DrugCLIP, a contrastive learning framework that reframes virtual screening as a dense retrieval problem—similar to how modern search engines work. The key innovation: encode protein pockets and small molecules into a shared latent space using separate neural networks, then use cosine similarity for ultrafast ranking. The model is pretrained on 5.5 million synthetic pocket-ligand pairs extracted from protein structures, then fine-tuned on 40,000 experimentally determined complexes. The speed gains are staggering—up to 10 million times faster than docking. Combined with GenPack, a generative module that refines pocket detection on AlphaFold2-predicted structures, DrugCLIP enables screening at a scale previously unthinkable: 500 million compounds against ~10,000 human proteins, scoring more than 10 trillion pairs in under 24 hours on just 8 GPUs. The wet-lab validations are equally compelling. For norepinephrine transporter (NET), a 15% hit rate with two inhibitors structurally confirmed by cryo-EM. For TRIP12—a challenging E3 ubiquitin ligase with no known inhibitors or holo structures—a 17.5% hit rate using only AlphaFold2 predictions, with functional enzymatic inhibition confirmed. The resulting database, GenomeScreenDB, covers ~20,000 pockets from 10,000 proteins—nearly half the human genome—and is freely available at drugclip.com. The message is clear: by combining contrastive representation learning with generative pocket refinement and AlphaFold structures, we've entered an era where genome-wide drug discovery becomes computationally tractable, opening systematic exploration of the vast undrugged proteome. Paper: https://lnkd.in/e7aGUvAX #DrugDiscovery #ArtificialIntelligence #MachineLearning #DeepLearning #VirtualScreening #ComputationalBiology #AlphaFold #ProteinScience #Biotech #AIforScience #StructuralBiology #Bioinformatics #Pharmaceuticals #ComputationalChemistry #PrecisionMedicine

💊 AI just made drug discovery searchable Virtual screening at genome scale has been computationally prohibitive. Traditional molecular docking works well for one target at a time, but screening large compound libraries against thousands of human proteins simultaneously would take years, even on modern GPU clusters. A team at Tsinghua University just changed this. They screened 500 million compounds against 10,000 human proteins, scoring 10 trillion protein-ligand pairs in under 24 hours using just 8 GPUs (!). Their new paper in Science (https://lnkd.in/ek5-d9F7) introduces DrugCLIP, a contrastive learning approach that's 10 million times faster than traditional docking. 🔬 How it works Two neural networks encode protein pockets and drug molecules into a shared embedding space, trained so that binders cluster near their targets while non-binders are pushed apart. Both encoders are built on UniMol, a 3D transformer that processes atomic coordinates directly rather than chemical formulas. The training is clever: pretrained on 5.5 million synthetic protein-fragment pairs, then fine-tuned on 44,000 real crystal structures using random conformations rather than exact poses - forcing the model to learn chemical features, not memorize geometry. Once trained, screening becomes nearest-neighbor search. 🚀 Why it's so fast The speed comes from pre-computation. You encode your 500 million molecules once and store the vectors offline. Screening a new protein target then becomes vector similarity - no physics simulations, no pose sampling, no energy minimization per molecule. 📊 The validation The team validated hits in wet-lab experiments. Traditional virtual screens typically yield 1-5% hit rates. DrugCLIP achieved: → 15% hit rate for norepinephrine transporter (NET), with structurally novel inhibitors distinct from existing drugs - two confirmed by cryo-EM → 17.5% hit rate for TRIP12, a target with no previously known ligands, using only AlphaFold-predicted structures That second result is remarkable - they found the first functional inhibitors for an unexplored target implicated in cancer and Parkinson's. 🌐 The resource The team released GenomeScreenDB (https://drugclip.com), an open-access database containing candidate molecules for ~20,000 pockets across ~10,000 human proteins - more targets than have any known ligands in ChEMBL. I think this represents a shift in how drug discovery will work. When screening becomes this fast and cheap, the bottleneck moves from computation to ideas: which targets matter, which patient populations to prioritize, how to validate hits efficiently. Congratulations to co-first authors Yinjun Jia, Bowen Gao, Jiaxin Tan, Jiqing Zheng, Xin Hong and senior authors Yanyan Lan, Wei Zhang, Chuangye Yan, and Lei Liu.

I'm pleased to share my latest article, "The State-of-the-Art in Machine Learning for Protein-Ligand Binding Affinity Prediction." In this overview, I discuss how modern machine learning models are being applied to predict drug–target affinity more efficiently. By examining the evolution from classical methods to advanced deep learning approaches, along with insights on data augmentation and model generalization, I explore both the progress made and the current challenges. The article offers a balanced view on how emerging techniques can narrow the gap toward more reliable, fast binding affinity predictions while still respecting the complexities of biochemical data. I invite you to read and share your thoughts on the practical implications of these developments for early-stage drug discovery💡. #drugdiscovery #CADD #computationalchemistry #ChemInformatics #AIindrugdesign #machinelearning #drugdesign #deeplearning #ArtificialIntelligence #AIinPharma

"Advances in covalent drug discovery" From their Abstract: Covalent drugs have been pivotal in disease treatment for over a century, with recent advances in tools for their rational design. Adding reactive functional groups to ligands enhances potent and selective inhibition of target proteins, exemplified by EGFR and BTK inhibitors for cancer. 'Electrophile-first' approaches have led to discoveries like KRAS(G12C) and SARS-CoV-2 protease inhibitors. Covalent screening technologies and chemoproteomics platforms now complement ligand discovery and profiling. This Review highlights milestones in covalent drug discovery, emphasizing lessons learned and the evolving toolbox driving success. Source: Boike, L., Henning, N.J. & Nomura, D.K. Advances in covalent drug discovery. Nat Rev Drug Discov 21, 881–898 (2022). https://lnkd.in/gJhKv_e7 #covalent_drugs #covalent_EGFR_inhibitors #covalent_ligands #drug_discovery

A new report “Beyond Legacy Tools: Defining Modern AI Drug Discovery for 2025 and Beyond,” is out! (link in the comments) The report by BiopharmaTrend (Disclaimer: I am a co-founder of the company) analyzes the AI platforms behind companies like Recursion, Insilico Medicine, Iambic Therapeutics, Schrödinger, Verge Genomics, NOETIK and several others — and shows that despite their different architectures and areas of focus, they share a set of category defining traits: ✔️ Modeling biology holistically, not just focusing on single targets or pathways ✔️ Building scalable, software-first platforms that integrate wet-lab and in silico workflows ✔️ Owning or generating massive, multimodal datasets (e.g. omics, imaging, patient data, and proprietary perturbation experiments) ✔️ Embedding AI at every stage of the pipeline, connecting the dots via gen AI. 👉 The report also introduces the concept of Holistic Drug Development (HDD), a vision where AI platforms integrate real-world patient data, systems biology, and generative chemistry into a continuous, learning-driven loop. Here is my take: “We’ve been using machine learning in biology for decades” is a common argument meant to downplay the idea that AI drug discovery (AIDD) is a new category. But IMO, this argument falls short. Yes, machine learning (ML) has been used in biology and chemistry for decades: QSAR models, clustering, PCA, support vector machines, and basic neural nets, etc. But those were point solutions: tools applied to narrow tasks (e.g., predicting solubility, docking ligands, or clustering gene expression data). What’s different now is that ML, particularly deep learning, generative modeling, and transformer architectures, is being used to rebuild the entire discovery workflows. Next, earlier ML approaches required handcrafted features (e.g., molecular descriptors). Today’s models can learn rich, abstract representations directly from raw data — from sequences, graphs, images, and text — and use them across tasks. That shift is foundational and category-defining. Also, traditional ML tools were modular and disconnected. Today’s AIDD platforms integrate multimodal data (omics, imaging, EHRs, chemical structures, etc). Modern AI drug discovery platforms, operate in closed feedback loops with wet-lab systems, offer full-stack software products with APIs, dashboards, and orchestration layers. That level of scope and systems integration is categorically different, IMO. Classical ML mostly focused on prediction and classification. AIDD platforms now generate novel chemistry, hypotheses, even trial designs, shifting from prediction tools to creative engines within the discovery process. Finally, the important aspect is production-grade software platforms, not just scripts and models. Using ML as a helper tool ≠ building AI-native, data-driven engines. I am pretty certain. Disagree? Image credit: BiopharmaTrend

In the realm of biotechnology, the quest for discovering novel therapeutic antibodies often resembles searching for a needle in a haystack. New research (https://lnkd.in/eiTnTYKg) from University of Oxford might just equip us with a super-powered magnet to make this search dramatically more efficient and precise. Context-Guided Diffusion (CGD), a novel AI method proposed by Leo Klarner, aims to transform how we predict and enhance the properties of proteins and molecules. Imagine trying to navigate a road that isn't yet mapped; CGD acts like an advanced GPS system, designed to navigate these uncharted territories in molecular landscapes. What Makes CGD Stand Out? CGD leverages unlabeled data and integrates them with smoothness constraints to improve the model's ability to predict new, effective molecular structures that were previously beyond our reach. For scientists and engineers in drug discovery, this means being able to innovate faster and more reliably, predicting molecular behavior that helps develop new treatments, and potentially reducing the time-consuming trial and error that typically characterizes laboratory experiments. The main limitation of CGD is its reliance on the availability of suitable unlabeled training data, which may not always be accessible or may require significant preprocessing. Open questions include how to optimize the selection and use of unlabeled data and how to extend the CGD framework to other types of biological data. Despite this, in tests across various domains, from small molecules to proteins, CGD consistently outperforms existing models, paving the way for it to become a powerful tool in accelerating therapeutic discoveries and bringing solutions to patients much quicker. Fantastic work by all the authors: Leo Klarner, Tim G. J. Rudner, Garrett Morris, Yee Why Teh, and the always impressive Charlotte Deane. #ArtificialIntelligence #MachineLearning #Biotechnology #DrugDiscovery #ProteinEngineering #Innovation

🔬 **Revolutionizing Drug Discovery: The Rise of Covalent Probes** 💡 Covalent probes are transforming the landscape of drug design, offering unprecedented opportunities to target "undruggable" proteins and achieve higher selectivity. Once dismissed due to concerns over toxicity and immunogenicity, these probes are now at the forefront of innovation, thanks to advancements in structure-based design and computational methods. A recent review in *Communications Chemistry* dives deep into the challenges and breakthroughs in designing covalent probes. Unlike non-covalent drugs, covalent probes require modeling multiple structural states—bound state, near-attack conformation, transition states, and product state—making their design far more complex. Yet, this complexity brings unique advantages, such as extended target inactivation and mutant-specific selectivity, exemplified by KRAS inhibitors like Sotorasib. The article highlights cutting-edge computational tools, from molecular dynamics to machine learning, that are enabling researchers to predict residue reactivity, optimize warhead chemistry, and model intricate reaction pathways. These innovations are paving the way for bespoke covalent chemistries tailored to specific targets, unlocking new possibilities in cancer treatment and beyond. As the field evolves, the integration of automated pipelines and novel ligation chemistries promises to push the boundaries of precision medicine. Covalent probes are no longer just a tool—they’re a paradigm shift. #DrugDiscovery #CovalentProbes #ComputationalChemistry #PrecisionMedicine #Innovation

Most drug discovery efforts still chase single targets. But complex diseases don’t work that way, they’re driven by networks, redundancy, and eerie adaptability. What if we could use AI, not to find “a hit,” but to shift entire cellular states from disease back toward health? A new Science paper (DeMeo et al., Oct 2025) takes that paradigm seriously. The team (Cellarity/MIT/Helmholtz) built DrugReflector, a deep-learning model trained on more than 1.2 million human cells and 88 chemical perturbations, using single-cell transcriptomics as its foundation. Instead of asking, “Will this molecule bind my target?”, they ask, “Will this molecule rewire the system toward a healthy phenotype?” What’s new here? DrugReflector predicts not just which compounds bind, but which will actually shift the transcriptomic signature of, say, a blood stem cell, into paths leading to functional megakaryocytes or erythrocytes, the cells we need for treating anemia or platelet disorders. They validated against brute-force screening (the industry standard): Random selection: ~1% hit rate Their model: up to 17%: a 13–17x improvement (and robust across several donors and pathways). The method uses closed-loop reinforcement learning. Initial predictions guide the first round of screening, then transcriptomic and phenotypic readouts from the real experiments refine the model in “lab-in-the-loop” cycles. With each iteration, the hit rate and biological insight both get sharper. It recovers known standards of care and highlights new targets. Notably, it identified both established kinase inhibitors and a new class of molecules modulating cholesterol synthesis to drive megakaryocyte commitment, finding druggable nodes unseen by classic screens. Why does this matter? Phenotypic drug discovery, with deep, biology-aware AI, can leapfrog the “screen everything” mentality, bringing tractability and true systems-level correction to disease treatment. Every cycle isn’t just screening, it’s learning: about cell fate, about target redundancy, and about network rewiring as therapy. The future lies in AI that understands and actively learns from biology, the cell as its own target, not just a test tube for single-protein hits. Are we ready to reimagine drug discovery workflows around this? The tools, and now, the evidence, are here!