I mainly use python for cheminformatics tasks but Javascript(JS) is required when I would like to web application. Because JS is powerful language for web app development. As many readers now that rdkit community provides rdkit-js. Link for npm package is below.https://www.npmjs.com/package/@rdkit/rdkit I’m not so good at modern JS such as Node, React, Vue and ..read more

Clustering librariries is one of the common but important task of cheminformatics. For example if you need to pick molecules from large amount of compound set, cluster these molecules and pick top X molecules will be one of the strategy. But in case of large compound set, clustering will be hard task ;) As you ..read more

As many readers know that DiffDock is one of the famous Deep Learning based docking program which use diffusion model. It was published 2022 at arxiv and code is disclosed on github. Recenlty I found that the repository was updated and code for web application is added :)https://github.com/gcorso/DiffDock I interested it because web app is ..read more

In the cheminformatics area we often use SMILES wich is 1D representation of molecules. But to conduct virtual screening we should prepare molecules from SMILES to 3D structure whith predict protonate state, enumerate chirality, tautomer and stereo isomer if necessary. There are lots of OSS libraries which can generate 3D conformers from SMILES such as openbabel and RDKit. Openbabel can generate protonate molecules from SMILES with user defined pH but RDKit can’t the option even if the 3D structure generation functoin work well. As many readers know that commercial packages can do it for examp ..read more

At hit to lead(H2L) stage in drug discovery projects, some attractive scaffolds are indentfied and explore SAR activity. And efficient method to select R-Groups from reagent list is requred as cheminformatics / CADD tasks. Selection requirements are variable, for example to get diversity, to keep exit vector and to get novelity etc. It depends on project situation. Lots of approaches are proposaled to tackle the challenge. And recently I read new and attractive approach to select building blocks from given reagent list. The proposal is publised from ACS JCIM by GSK’s team. The link is below. b ..read more

If target structure of the drug discovery project is available, SBDD is one of the practical approach. Docking study is common method of SBDD. There are lots of docking simulation softwares recently not only commercial packages but also open source software. AutoDock Vina is one of the famous Docking tool in the area. To run the docking study, data preparation from receptor and ligands is required. For example in ligand preparation, several steps are requred, 3D conformation generation, predict protonation state, enumerate tautomer (if requred). After preparation docking study is conducted wit ..read more

As many readers know that currently GPU based docking for massive number of comound is trend. Because speed of GPU Docking is faster than CPU based. Of course if you can use huge amount of CUPs, CPU docking is aslo fast ;) AutoDock-Vina is one of the famous OSS docking tool. And GPU version is published in 2023. The article shows that good correlation between CPU and GPU based vina score. I would like to chack it on my PC. Fortunately UniDock provides example dataset. So I used it and tried to comparison. Following experiments, I used Vina-GPU 2.1 which is the latest version of Vina-GPU. And I ..read more

As many cheminformaticians know that Openbabel is one of the famous and useful package of Cheminformatics as same as RDKit. Openbabel provides not only CLI but also API for some programming languages including python of course ;). Openbabel can protonate molecule with user defiened pH. The function is not available from current version of RDKit. By using obabel script with -p (pH) option I could protonate molecules but I would like to know how to do samething from python API. After struggling coding I found the way to do it. Here is an example. from openbabel import openbabel ob obc = ob.OB ..read more

Now we can use folding AI such as Alpha-fold to predict 3D structure of target proteins of course homology modeling is also used to do it. But these structures are snap shot and apo form. So Ligand based drug design(LBDD) Ligand based virtual screening (LBVS) are still important strategy for drug discovery. Lots of LBVS is based on compound similarity which means similar compounds will have similar biological properties. The metrics of similarity is used not only 2D but also 3D. Today I would like to try to use the LBVS package which is developed by Healx, AI based drug discovery company. They ..read more

As name of LLM means that to use these kinds of models, we need enough GPU memory and it’s not so cost effective for personal use ;) To overcome the limitation, there are lots of technologies are developt and still be developping. LLAMA-cpp is one of the them. Today I would like to share new model which is named phi3 developed by microsoft. The original article is found in arixv. https://arxiv.org/pdf/2404.14219 From the abstract, “phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, whose overall performance, as measured by both academic benchmarks and internal t ..read more