Shahriyar Gourgi’s Post

Antibody therapeutics are highly specific and powerful at stopping disease... if they are precisely engineered. A-Alpha Bio's newly-published AlphaBind optimization model does it 𝘧𝘢𝘴𝘵𝘦𝘳, 𝘤𝘩𝘦𝘢𝘱𝘦𝘳, and 𝘣𝘦𝘵𝘵𝘦𝘳. 𝐍𝐞𝐰𝐬 𝐚𝐭 𝐚 𝐠𝐥𝐚𝐧𝐜𝐞: 💊 Leveraging the body's own immune system, antibody therapies offer a versatile, modular, and precise solution to treating hard-to-target diseases. But the cost and time required to produce the 𝘳𝘪𝘨𝘩𝘵 ones—those with low immunogenicity and long-lasting effects—get in the way of innovation. 🧬 A-Alpha Bio's AlphaBind antibody optimization computational model antibody-antigen binding data to engineer effective therapeutics in a fraction of the time it would take in a traditional biopharma lab. 💻 Built on top of NVIDIA's BioNemo platform, AlphaBind is now publicly available on Github, along with the AlphaSeq datasets that can help fine-tune it for your own research needs: https://lnkd.in/dz_8Uce8 𝐑𝐞𝐚𝐝 𝐭𝐡𝐞 𝐟𝐮𝐥𝐥 bioRxiv & medRxiv 𝐚𝐫𝐭𝐢𝐜𝐥𝐞: https://lnkd.in/gT-96Va8

H3-OPT: Overcoming Antibody CDR-H3 Structure Prediction Challenges with Deep Learning 📢 Chen et al. recently reported H3-OPT, a model for predicting structures of the heavy chain 3 of antibody (Ab) complementarity-determining regions (CDR-H3) based on AlphaFold2 (AF2). For both monoclonal Abs (having both heavy and light chains) and nanobodies (with a single-domain heavy chain), the CDR-H3 loop plays a key role in antigen binding and is thus the most diverse Ab region in terms of amino acid length and composition. 💉 Given the importance of the CDR-H3, accurate development of Ab-based therapeutics is dependent on experimental structures of candidate Abs, a process which is bottlenecked by both cost and time. Thus, accurate computational prediction methods would facilitate the therapeutic Ab development pipeline. However, existing methods struggle to generate high quality predictions of the CDR-H3 loop regions because of the inherent challenge of predicting loop structures. 🔥 H3-OPT integrates the strengths of AF2 with a pre-trained protein language model (PLM) to predict CDR-H3 structures. Based on the observation that AF2-predicted Ab structures show an overall high quality, H3-OPT extracts the structural features and uses the information to generate refined CDR-H3 structures. H3-OPT consists of two modules: a template module and a PLM-based structure prediction module (PSPM). The template module dictates whether the PSPM is deployed for a particular query. The PSPM is further comprised of two sub-modules: a confidence-based module that evaluates the quality of the CDR-H3 from AF2, and a template-grafting module which identifies a suitable PDB template to replace low quality AF2 H3 loop. 🔢 On a benchmark dataset, H3-OPT outperforms other state-of-the-art methods, including AF2, IgFold, HelixFold-Single, ESMFold, and OmegaFold, by achieving an RMSD (to experimental structure) of 2.24 Å, compared to 2.85 Å and 2.87 Å for the next best methods, AF2 and IgFold, respectively. Furthermore, H3-OPT was evaluated using recently deposited PDB structures of three anti-VEGF variants, giving predictions with RMSDs of 1.510 Å, 1.541 Å, and 1.411 Å for the three variants. Predictions by IgFold, however, gave RMSDs of 2.776 Å, 2.888 Å, and 2.448 Å. 💪 H3-OPT also exhibited improved performance on other tasks, such as the prediction of Ab CDR-H3 surface properties (including surface amino acids, solvent-accessible surface areas, and surface charge distribution) and antibody-antigen interactions. Paper: https://lnkd.in/gTRENbHR GitHub: https://lnkd.in/g43BNRRw #structuralbiology #biologics #ML #biopharmaceuticals #genophore

Combinatorial prediction of therapeutic perturbations using causally-inspired neural networks ABSTRACT As an alternative to target-driven drug discovery, phenotype-driven approaches identify compounds that counteract the overall disease effects by analyzing phenotypic signatures. Our study introduces a novel approach to this field, aiming to expand the search space for new therapeutic agents. We introduce PDGrapher, a causally-inspired graph neural network (GNN) designed to predict combinatorial perturbagens - sets of therapeutic targets - capable of reversing disease effects. Unlike methods that learn responses to perturbations, PDGrapher solves the inverse problem, which is to infer the perturbagens necessary to achieve a specific response - i.e., directly predicting perturbagens by learning which perturbations elicit a desired response. By encoding gene regulatory networks or protein-protein interactions, PDGrapher can predict unseen chemical or genetic perturbagens, aiding in the discovery of novel drugs or therapeutic targets. Experiments across nine cell lines with chemical perturbations show that PDGrapher successfully predicted effective perturbagens in up to 13.33% additional test samples and ranked therapeutic targets up to 35% higher than the competing methods, and the method shows competitive performance across ten genetic perturbation datasets. A key innovation of PDGrapher is its direct prediction capability, which contrasts with the indirect, computationally intensive models traditionally used in phenotype-driven drug discovery that only predict changes in phenotypes due to perturbations. The direct approach enables PDGrapher to train up to 25 times faster than methods like scGEN and CellOT, representing a considerable leap in efficiency. Our results suggest that PDGrapher can advance phenotype-driven drug discovery, offering a fast and comprehensive approach to identifying therapeutically useful perturbations. PAPER: https://lnkd.in/d85Vw-gB

Antibody Design Enters the AI Era Traditionally, this process has been a slow and laborious affair, involving screening vast numbers of antibody candidates and often failing to yield drugs with the desired characteristics. But now, artificial intelligence (AI) is emerging as a powerful tool that promises to accelerate and improve this critical process. In my opinion, the biggest challenge in traditional antibody discovery is its inefficiency. Screening massive libraries of candidate antibodies is time-consuming and expensive, and it doesn't always guarantee success. Researchers often struggle to produce antibodies with the necessary properties for effective drugs. This is where AI comes in. Biotech companies are leveraging machine learning to streamline antibody discovery. These AI platforms analyse vast datasets and use this information to design and optimise antibody candidates, all while integrating with wet lab capabilities to validate their designs. Faster Discovery AI can rapidly analyse and optimise antibody candidates, significantly reducing the time it takes to identify promising drug leads. Improved Specificity Machine learning can design antibodies with a higher degree of specificity, meaning they target only the disease-causing cells and minimise side effects. De Novo Design Looking ahead, AI has the potential to design entirely new antibodies from scratch, ushering in a new era of drug discovery possibilities. The implications of AI-powered antibody design are vast. Faster development timelines mean drugs can reach patients sooner. Improved specificity translates to safer and more effective treatments. And the ability to design entirely new antibodies opens doors to tackling previously undruggable targets. This is just the beginning of the AI revolution in antibody design. The recent high-profile launch of Xaira, an AI-focused biotech with over $1 billion in funding, underscores the immense potential of this field. As AI capabilities continue to advance, we can expect to see even more groundbreaking discoveries emerge. What would you add? Found this useful? Repost ♻️ to help your network. Join a community of 85,600+ HealthTech leaders finding the ideas, people, innovations and technologies that are shaping the future of healthcare. 👉 http://lnkd.in/eExMcaG6

Revolutionizing Medicine, One Antibody at a Time. Imagine a future where diseases are not just managed, but cured, thanks to precision-engineered antibody drugs. It's not just a dream. It's becoming a reality, and AI is at the heart of this transformation. Scientists are leveraging AI to optimize the development of antibody drugs, pushing the boundaries of what's possible in medicine. By iteratively mutating amino acids, AI algorithms are pinpointing the most effective sequences for targeting diseases. It's a game-changer. This approach isn't just about speeding up drug development. It's about precision, efficiency, and the potential to save millions of lives by creating more effective therapies. As a CEO deeply entrenched in the AI space, I see this as a monumental step forward. It validates the power of generative AI in solving complex biological challenges. The implications for healthcare are profound, offering a glimpse into a future where our fight against diseases is bolstered by AI's unparalleled capabilities. For those of us at the intersection of AI and healthcare, this is a moment of validation and anticipation. We're not just watching the future unfold; we're helping to shape it. Read more about this groundbreaking work and its implications for the future of medicine: [Link to the article] The fusion of AI and biology is not just the next chapter in healthcare; it's a whole new book. Let's turn the page together. Check this out: https://lnkd.in/gUKdpkBP

IgBlend: Merging 3D Structures and Sequences for Enhanced Antibody Engineering The integration of machine learning into antibody discovery workflows is poised to massively impact therapeutics discovery and development. IgBlend aims to unify 3D structures and sequences in antibody language models. This advancement addresses a critical gap in current large language models (LLMs) for antibodies. The Need for Integration of Structure and Sequence Traditional LLMs trained solely on antibody sequences have made significant strides in antibody engineering. However, these models often overlook the structural information. Structural data provides a richer, more informative context that can significantly enhance the learning capabilities of language models. IgBlend leverages both sequence and structural information to better understand and predict the functional properties of antibodies. A Multi-Modal Approach Trained on an extensive and diverse dataset comprising over 4 million unique structures and more than 200 million unique sequences—including heavy and light chains as well as nanobodies. Performance Highlights Sequence Recovery: 👉 IgBlend achieved 83.80% accuracy in heavy chains, 80.07% in light chains, and 37.37% in nanobodies. 👉 Inverse folding models like AntiFold and ESM-IF showed lower accuracies, with IgBlend outperforming them by significant margins. CDR Editing: 👉 IgBlend (Seq + Struct Guided) demonstrated improvements of up to 15.43% in accuracy for nanobodies compared to sequence-only models. 👉 For the heavy chain CDR3 (H-CDR3), IgBlend showed a 53.35% accuracy compared to AntiFold's 36.27%. Inverse Folding: 👉 IgBlend consistently ranked highest in generating sequences that fold back to the original structure, outperforming AntiFold and ESM-IF across all temperatures and chain types. 👉 For example, at temperature T=1e-4, IgBlend achieved 73.60% RMSD <1Å for heavy chains, 97.80% for light chains, and 81.00% for nanobodies, compared to AntiFold's 63.60%, 89.20%, and 43.00% respectively. HER2 H-CDR3 Design: 👉 IgBlend (inverse folding) showed a strong positive correlation between binding affinity and scoring, with Kendall’s τ of 0.15 and Spearman’s ρ of 0.23, outperforming sequence-only models like AbLang2 which had Kendall’s τ of 0.04 and Spearman’s ρ of 0.07. Sequence Diversity Issues While IgBlend marks a substantial improvement over existing models, it does exhibit a trade-off between sequence diversity and accuracy. Specifically, incorporating structural information sometimes results in lower Levenshtein distances, indicating reduced sequence diversity. Future work should aim at incorporate side-chain information and expand structural datasets to further enhance the model's capabilities, and expand the model to handle paired sequences more effectively. #Machinelearning #Drugdiscovery #Biologics Paper: https://lnkd.in/gyneHuC9 Code: unpublished

IgGM: A Generative Model for Functional Antibody and Nanobody Design ABSTRACT Immunoglobulins are crucial proteins produced by the immune system to identify and bind to foreign substances, playing an essential role in shielding organisms from infections and diseases. Designing specific antibodies opens new pathways for disease treatment. With the rise of deep learning, AI-driven drug design has become possible, leading to several methods for antibody design. However, many of these approaches require additional conditions that differ from real-world scenarios, making it challenging to incorporate them into existing antibody design processes. Here, we introduce IgGM, generative model that combines a diffusion model and the consistency model for generating antibodies with functional specificity. IgGM produces antibody sequences and structures simultaneously for a given antigen, consisting of three core components: a pre-trained language model for extracting sequence features, a feature learning module for identifying pertinent features, and a prediction module that outputs designed antibody sequences and the predicted complete antibody-antigen complex structure. IgGM has shown effectiveness in both predicting structures and designing novel antibodies and nanobodies, making it relevant in various practical scenarios of antibody and nanobody design. PAPER: https://lnkd.in/ddBFc92A CODE: https://lnkd.in/d9sbNbJA

ImmuFold: High-Accuracy Antibody Structure Prediction with Efficient Network ABSTRACT Antibody structure prediction is a critical task in immunological research and therapeutic antibody development. Despite advances in prediction methods, contemporary approaches still face formidable challenges, particularly in accurately modeling Complementarity-determining regions (CDRs). Furthermore, current prediction time costs, typically on the order of minutes, preclude large-scale structure prediction and screening. In this work, we present ImmuFold, a novel deep-learning approach that achieves second-level performance in antibody structure prediction. ImmuFold integrates ImmuBERT, a 650M antibody language model pre-trained on hundreds of millions of natural antibody sequences, with a structure prediction network that directly predicts all-atom structure, encompassing both main chain and side chains. ImmuFold outperforms current methods, including IgFold and AlphaFold2, generating higher-quality antibody structures in approximately one second. Comparative analysis of the antibody binding task demonstrates the superior representational capabilities of ImmuBERT relative to existing language models, a crucial factor underpinning the efficacy of ImmuFold. PAPER: https://lnkd.in/dWVZ_jAw

Fast and accurate modeling and design of antibody-antigen complex using tFold Abstract: In this study, we present a fully end-to-end approach for three-dimensional (3D) atomic-level structure predictions of antibodies and antibody-antigen complexes, referred to as tFold-Ab and tFold-Ag, respectively. tFold leverages a large protein language model to extract both intra-chain and inter-chain residue-residue contact information, as well as evolutionary relationships, avoiding the time-consuming multiple sequence alignment (MSA) search. Combined with specially designed modules such as the AI-driven flexible docking module, it achieves superior performance and significantly enhanced speed in predicting both antibody (1.6% RMSD reduction in the CDR-H3 region, thousand times faster) and antibody-antigen complex structures (37% increase in DockQ score, over 10 times faster), compared to AlphaFold-Multimer. Given the performance and speed advantages, we further extend the capability of tFold for structure-based virtual screening of binding antibodies, as well as de novo co-design of both structure and sequence for therapeutic antibodies. The experiment results demonstrate the potential of tFold as a high-throughput tool to enhance processes involved in these tasks. To facilitate public access, we release code and offer a web service for antibody and antigen-antibody complex structure prediction, which is available at https://lnkd.in/dDKrX88G. Paper: https://lnkd.in/d6vJ9kpy Code