Hello Everyone I am Manan Goel, an undergraduate student at IIIT Hyderabad (India) and I'll be contributing to RDkit for GSoC 2020. My project involves integrating trained neural networks with RDkit<https://summerofcode.withgoogle.com/projects/#5947659405230080>. I aim to add the ANI neural network potential for molecular property prediction which uses the Behler Parrinello molecule featurization scheme to RDKit because it provides accuracy comparable to Quantum Mechanical Approaches like Density Functional Theory / MD / MC but do the job in a fraction of the time. The first part of the project would involve implementing the Behler Parrinello featurization scheme following which I will add the ANI1-cc model which takes atomic environment vectors as input for energy predictions. The authors of the ANI approach have provided pre-trained weights for energy prediction of molecules so the next part would be to create an interface to load these pre-trained weights to the model. The next part would be to implement back-propagation and find derivatives of the energy with respect to atomic positions to get forces on each atom which can then be used for molecular conformation optimization. It's amazing to be a part of the community and I hope to learn a lot from the experience. Regards Manan
_______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
