Nostrum Biodiscovery

We're excited to introduce GeoDirDock (GDD), our new diffusion model for directed docking. Diffusion models often overlook that the docking site on a protein is typically predetermined, along with specific molecular interactions at that site. GDD leverages this information, employing a directed diffusion process that guides docking to the desired pose through geodesic paths in implicit spaces. Our approach not only surpasses other rigid diffusion models like DiffDock in PDBBind standard benchmarks but also generates more accurate molecular poses. These poses show fewer clashes and maintain better proximity to side chains and backbones, enhancing the plausibility and effectiveness of the docking predictions. See in the examples below how GDD (green) it's able to find the "right" pocket to bind (crystal ligand in red) while DiffDock (grey) explores a cavity that is not subject to the screening. Also, for large molecules, the imposed prior it's able to fix most of the torsional angles and sets GDD to get the pose right. Even challenging for traditional docking methods! Stay tuned for the upcoming release of GDD, which will be made accessible to all users soon. This work has been accepted at the Generative and Experimental Perspectives for Biomolecular Design Workshop at ICLR. You can check out the paper here: https://lnkd.in/dZF5ywnh . Feel free to reach out to our AI Team and its authors (Raúl Miñán Campos, Javier Gallardo, Alvaro Ciudad, Alexis Molina) for further questions! #technology #GeoDirDock #Biotech #Pharma #Bioinformatics

Secure your place at the highly anticipated Poster Presentation on accelerating antibody design using cutting-edge AI+physics-based computational frameworks. 🤖⚛️🔬 Book your place for free! https://lnkd.in/en8-Pt2N Key topics include: 🔹 Harnessing AI and physics-based techniques to explore antibody-antigen interactions 🔹 Understanding binding modes and estimating affinity 🔹 Engineering more potent antibodies for therapeutic use 🧪💡 Presented by Sergi Rodà Llordés, Director of Protein Engineering at Nostrum Biodiscovery who will showcase ground-breaking innovations in antibody discovery and engineering. 🧑🔬🧠 📍 Register now to secure your spot and stay at the forefront of antibody development! 💻📊 📅 18 Nov 2024 🕒 16:30 - 17:00 🎯 Track: Lead Identification & Optimization 📍Antibody Therapeutics Xchange - Brussels 2024 Only available for Senior Scientists and above, from Bio and Pharma companies, with a drug pipeline. 💊 #AIinBiology #AntibodyEngineering #ComputationalBiology #ProteinEngineering #DrugDiscovery #AntibodyTherapeutics #GenerativeAI #Biotech #Pharma #MedicalResearch #LifeSciences #Innovation #ScientificRoundtable #hubXchange #Xchanges #Roundtable #FreePass #RegisterNow #ComplimentaryPass

We're #hiring!!! 🔍 At Nostrum Biodiscovery we are seeking a new Computational Protein Engineer for the Protein Engineering area. ¡We're waiting for you! ⬇⬇ ⬇ ⬇ ⬇ ⬇ ⬇ ⬇

We're #hiring!!! 🔍 At Nostrum Biodiscovery we are seeking a new Biomolecular Modeler for the Drug Discovery area. ¡We're waiting for you! ⬇⬇ ⬇ ⬇ ⬇ ⬇ ⬇ ⬇

Being part of Almirall R&D highlights makes us extremely proud. Our entire team at Nostrum Biodiscovery is dedicated to enhancing the finest outcomes of artificial intelligence-powered de novo design of molecules.

XNA-hub: the first global comprehensive database and molecular modeling platform for XNA research, to make it more efficient. To facilitate the computational investigation of XNAs, XNA-hub seeks to centralize all of the parameters (FF-components) involved, generate 3D single-stranded (ssXNA) and 3D double-stranded (dsXNA) structures automatically based on sequence demand, and predict thermodynamic properties of XNAs and their hybrid derivatives by comparing computational observations with experimental data gathered from web scraping. Lastly, by heavily utilizing machine learning techniques, XNA-hub provides predictions for RNAseH compatibility (ASOs) and AGO2 tolerance (for siRNA). #xna #research

We are incredibly fortunate to have Adrián Morales Pastor as a part of our team, and we are grateful for the opportunity to count on his talent and contributions. As we celebrate his achievements and dedication, we look forward to many more years of collaboration and success together. Thank you for your outstanding work and for being an invaluable team member. #ai #artificialintelligence #drugdiscovery

I am thrilled to share that I have had the opportunity to attend and participate in the Forty-first International Conference on Machine Learning (ICML 2024) held at the Messe Wien Exhibition Congress Center in Vienna, Austria. It was an incredible week filled with insightful talks, innovative research presentations, and engaging discussions on the latest advancements in machine learning. A special highlight of the conference for me was the GRaM Workshop on Geometry-grounded Representation Learning and Generative Modeling. Our team from Nostrum Biodiscovery and BSC, including Alexis Molina, Bertran Miquel Oliver, and myself, had the honor of participating in the ICML Topological Deep Learning Challenge 2024: Beyond the Graph Domain. We were awarded as great contributors and are excited to have the opportunity to extend our work into a paper. I would like to extend my heartfelt thanks to the organizers of the GRaM Workshop and the ICML conference. Looking forward to applying the knowledge and insights gained from this experience to our ongoing projects at Nostrum Biodiscovery! #ICML2024 #MachineLearning #GeometricDeepLearning #TopologicalDeepLearning

Time to present our second work accepted to the AccMLBio Workshop at [ICML] Int'l Conference on Machine Learning'24 where we posed the following question: Are protein language models (pLMs) compute optimal? In our research, we leveraged our own models and embeddings for downstream tasks. It is crucial for us to balance model size, performance, and resource utilization. In this study, we investigated how the scaling laws derived in NLP, as described by Kaplan et al. and Hoffman et al., apply to the scaling of pLMs and their compute optimality. Our findings revealed that when fitting on UniRef subsets built at different identity cutoffs, which is a commonly used sequence set, ESM-like models plateau at a loss of 2.4 regardless of model size. Moreover, while pLMs are typically trained for several epochs, we observed that less than a single pass over the dataset is sufficient to achieve highly competitive performance with models such as ESM-2 (15B) and xTrimoPGLM (100B). Notably, our model achieved these results after being trained for just one hour on a single NVIDIA H100 GPU!!! These results demonstrate that foundational pLMs can be made more accessible to SMEs and academic labs by implementing a more rational training approach. Stay tuned for more work to be released in the following months! You can check the paper out here: https://lnkd.in/d_Xxq8rq Also, if you are in Vienna for the ICML'24 workshops, do not hesitate to reach out to our authors Yaiza Serrano, Alvaro Ciudad, and Alexis Molina, who will be attending the event. #computational #ai #artificialintelligence

The Newsletter of Q2 2024 ⭐ Stay tuned for upcoming releases