SMILES-X

The SMILES-X is an autonomous pipeline that finds best neural architectures to predict a physicochemical property from molecular SMILES only (see OpenSMILES). No human-engineered descriptors are needed. The SMILES-X has been especially designed for small datasets (<< 1000 samples).

Read our open access paper "SMILES-X: autonomous molecular compounds characterization for small datasets without descriptors".

Who can use the SMILES-X?

Researchers/engineers/students in the fields of materials science, physicochemistry, drug discovery and related fields

Which kind of data can be used?

The SMILES-X is dedicated to small datasets (<< 1000 samples) of (molecular SMILES, experimental/simulated property)

What can I do with it?

With the SMILES-X, you can:

• Design specific neural architectures fitted to your small dataset via Bayesian optimization.
• Predict molecular properties of a list of SMILES based on designed models ensembling without human-engineered descriptors.
• Interpret a prediction by visualizing the salient elements and/or substructures most related to a property

What is the efficiency of the SMILES-X on benchmark datasets?

• ESOL: logarithmic aqueous solubility (mols/L) for 1128 organic small molecules.
• FreeSolv: calculated and experimental hydration free energies (kcal/mol) for 642 small neutral molecules in water.
• Lipophilicity: experimental data on octanol/water distribution coefficient (logD at pH 7.4) for 4200 compounds.

All these datasets are available in the data/ directory above.

Requirements

SMILES-X works on CPU, but it is highly recommended to have an access to GPU:

• CUDA == 10.1
• cuDNN == 8.0.3

For now, the SMILES-X has been successfully runned on Titan(Xp, V, V100, P100), GTX 1660 and RTX 2070/80 NVIDIA GPUs.

For a good start, follow the RDKit installation guide for installing the RDKit via conda.
Then, install the following dependencies in your RDKit conda environment (e.g. my-rdkit-env):

• python == 3.7
• pandas == 1.3.5
• numpy == 1.18.5
• matplotlib == 3.5.2
• GPy == 1.10.0
• GPyOpt == 1.2.6
• scikit-learn == 1.0.2
• adjustText == 0.8
• scipy == 1.7.3
• tensorflow == 2.1.3

Usage

Please use the python notebook Example.ipynb it as a guide.

How to cite the SMILES-X?

@article{lambard2020smiles,
  title={SMILES-X: autonomous molecular compounds characterization for small datasets without descriptors},
  author={Lambard, Guillaume and Gracheva, Ekaterina},
  journal={Machine Learning: Science and Technology},
  volume={1},
  number={2},
  pages={025004},
  year={2020},
  publisher={IOP Publishing}
}