Sequence-based preprocessors work with atoms

[ayushkarnawat]

Should sequence-based preprocessors compute features for sequences that contain atoms as well as residues? Currently, only features are computed for sequences that contain residues.

This would make sense if using a SDF file to read molecules. Since .sdf files are only able to read atom-level information (they have no concept of identifying residues unlike PDB files), it could allow us to compute sequence-based features (using sequence-based preprocessors) for SDF files rather than just molecular 3D features (using mol-based preprocessors).

Note: We will also have to think about how to process pre-trained atom-level embeddings along with protein-embeddings (see #5). Not directly related, but good to think about are mutations with the SDF files (#10) and mutating arbitrary sequences (#7).