Yinghong Gao’s Post

Molecular Docking: Key Concepts and Applications • #Molecular #docking is a computational technique used in structural #biology and #drug #discovery to predict interactions between a small molecule (ligand) and a receptor target (protein). • Key concepts include: - #Ligand and #Receptor: Understanding how a small molecule interacts with a larger receptor molecule is crucial for drug design and development. - #Docking #Algorithms: Various algorithms predict the optimal binding pose of the ligand within the receptor binding site. - Scoring Functions: These functions assess the strength of the ligand-receptor interaction, considering factors like electrostatic interactions, van der Waals forces, hydrogen bonding, and desolvation energies. • Applications include drug discovery, virtual screening, and understanding #biological processes. • Challenges include scoring accuracy, flexibility, computational cost, and integration with experimental data. • Future directions include combining computational predictions with experimental data, applying machine learning techniques, and developing algorithms that account for molecular flexibility. #moleculardocking #structuralbiology #proteindatabase #proteins #ligands #analysis #analyticaltechniques #linkedin #moleculardockingstudy

Unlocking new therapeutic potentials through advanced computational techniques and AI-driven molecular modeling. It is a wonderful review to dive deep into the resources for virtual screening. #DrugDiscovery #Cheminformatics #VirtualScreening #AIdrivenVS

Lipid Nanoparticles Efficiently Deliver the Base Editor ABE8e for COL7A1 Correction in Dystrophic Epidermolysis Bullosa Fibroblasts In Vitro https://lnkd.in/e7YXaiSJ #JIDJournal #dermscience

Move over, ligand RMSD < 2 Å! ConfBench is on the scene! If you're interested in the evaluation of conformational accuracy of structure prediction methods, take a look at our first stab at a systematic conformational benchmark in the NP3 technical report below ↓ ConfBench systematically evaluates ligand-induced conformational changes across the proteome by comparing predictions against reference apo/holo states. it focuses on pdb pairs with measurable differences (>1.5Å rmsd) in global structure or ligand binding pocket environment. Using ConfBench, we found NP3 outperforms AF by a 20% margin on recent structures, while maintaining consistent prediction success rates across both apo and holo targets - suggesting NP's capability for robust conformational predictions. Ultimately, current structure prediction models don't truly understand physics - they learn patterns from end-state structures. Most models pretty consistently predict holo states, essentially hallucinating ligands due to training data skew (as the majority of the PDB is holo). There's no point in trying to get ML models to perfectly "learn" physics - we'd just end up spending the same compute as expensive physics-based models! Instead, what data signals could serve as meaningful proxies for physics-driven phenomena? While our physics-based priors have improved NP3's performance on apo/holo predictions, we're likely reaching the limits of what can be learned from static structures alone. the field needs to consider new data sources to advance further. Potential directions include physics-based synthetic data generation through molecular dynamics simulations, quantum mechanical calculations, and physics-informed neural networks. each approach offers unique insights into protein dynamics that complement existing structural data. These synthetic data sources could help bridge the gap between pattern recognition and physical understanding, particularly for challenging cases like predicting full protein conformational landscapes where experimental data is scarce. tl;dr... ConfBench provides a quantitative framework for measuring progress in conformational prediction. We've exhausted traditional benchmarks - it's time for a paradigm shift in how we evaluate and improve protein structure prediction methods for drug discovery. Unimaginably large thank you to the amazing team and leadership that made this work possible, especially Zhuoran Qiao and Matthew Welborn for believing in my vision of conformational enablement of structure prediction, and to my PhD advisor Rommie Amaro who sparked my love for dynamic proteins.

OpenEye's advanced CSP workflow offers a well-validated approach to predicting the crystal form landscape of a molecule, helping identify alternative crystal forms (polymorphs) early in the drug development process. Join us for our upcoming webinar to explore the latest innovations in this workflow!

We are pleased to release today the technical report for NeuralPlexer3 (NP3), our new state-of-the-art generative model for protein-ligand structure prediction. NP3 advances the field of AI-driven drug discovery, providing researchers precise detail on small molecule/protein interactions. NP3 also provides greater speed and prediction accuracy than existing models on all biomolecular interaction types (proteins, nucleic acids, ligands, ions, and post-translational modifications). In a fast moving field, we feel that providing a transparent mechanism for head-to-head comparisons is essential for meaningful scientific progress. To that end, today we also shared code for our primary benchmarking approach on GitHub. By supporting target validation, structural refinement, and atom-level interaction analysis, NP3 bridges critical gaps in AI-driven drug discovery, empowering researchers to navigate the complexities of therapeutic development with greater precision and speed to accelerate therapeutic innovation. https://lnkd.in/gQf4iikb https://lnkd.in/gf3uy_SM #AI #GenAI #BioTech #Innovation #NeuralPLexer

Today we are releasing NeuralPLexer3 (NP3) technical report, along with opening-sourcing NPBench https://t.co/y5xZ9VXtEc - our turnkey library for fair benchmarking of co-folding structure prediction models on diverse interactions - a task that has challenged this field. NP3 adopts a flow-matching framework with optimal transport to handle intrinsic symmetries, pushing the accuracy frontier on Posebusters-v2 (77.9% RMSD<2A & PB-valid) while being up to 15x faster than AlphaFold 3 (AF3). Enabled by NPBench, NP3 shows better or competitive performance to AF2-M and AF3 on a comprehensive set of 1,143 post-2022, low-homology, arbitrary stoichiometry PDB targets. It is also excellent at confidence estimation (pLDDT - ligand RMSD, pDockQ - DockQ). We present model scaling studies with full LDDT breakdown and ablation results obtained throughout the entire course of model development, highlighting a compute-optimal frontier of encoder and decoder capacity and the importance of inference-time improvements. Our Flash-TriangularAttention kernel dramatically reduces peak memory usage by using tiling to eliminate unnecessary bias replication, while supporting gradient-tracked pair bias that are important in modern structure prediction model backbones. But being accurate and speedy on structure prediction benchmarks is only halfway towards the goal. At Iambic, we are pursuing active research to better deliver these tools to real-world drug discovery usage cases. We introduce ConfBench, a robust scoring protocol to evaluate model prediction on ligand-induced protein conformational changes. On recent structures NP3 outperforms AF2-M with a 20% margin, maintaining similar conformational prediction success rates on apo and holo targets. Huge thanks to the contribution from our fantastic colleagues: Feizhi Ding (NPBench), Thomas Dresselhaus (data), Mia A. Rosenfeld, PhD (ConfBench), Xiaotian (Max) Han (Triton kernels), Owen Howell, Stephen Opalenski, Aniketh Iyengar, Anders Christensen, Sai Krishna S., and the tremendous support from our leadership team (Thomas Miller, Fred Manby, Matthew Welborn)! If you are interested, please give NPBench a try - we advocate all researchers and leading players in the space to share their model predictions, new datasets, and hosting infrastructures to identify new opportunities for structure prediction and advance the field further.

Accurate structure prediction of biomolecular interactions with AlphaFold 3 Abstract The introduction of AlphaFold 21 has spurred a revolution in modelling the structure of proteins and their interactions, enabling a huge range of applications in protein modelling and design. Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues. The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer. Together, these results show that high-accuracy modelling across biomolecular space is possible within a single unified deep-learning framework. https://lnkd.in/dxX4E9_B

Discover the power of untargeted lipidomics research with LC-MS/MS technology! 🔬 Maximize your lipid analysis with high-resolution mass spectrometry, identifying changes in lipid composition across biological processes. Harness the advantage of 1.7M lipid molecules mapping! #LipidomicsResearch #MassSpectrometry https://lnkd.in/gyVSEiFm

🔬 Happy to share our latest publication! 💊We combined #machinelearning and #molecularsimulations to unveil the free energy landscapes of aminosterol absorption. This study, published in the Journal of Chemical Theory and Computation, offers crucial insights for targeted drug delivery and neurodegenerative disorder treatments. Key highlights: ✅ Leveraged Deep-TICA approach to analyze membrane insertion of trodusquemine and squalamine ✅ Developed an effective method to describe solute absorption and estimate free energy landscapes ✅ Achieved accurate predictions of membrane binding affinities, validated by experimental data ✅ Deepened understanding of aminosterol-lipid membrane interactions Read the full paper: https://lnkd.in/eup7DVTV #drugdiscovery #biophysics #computationalchemistry