DeepChem Papers and Discoveries List

Paper Title: DGL-LifeSci: An Open-Source Toolkit for Deep Learning on Graphs in Life Science
Summary of DeepChem Usage: MoleculeNet and DeepChem models (non DGL) are used as baselines.
Important Contributions: Major improvements in performance of graph conv models.
Date Published: June 2021
Journal: Arxiv

Paper Title: Chemical Toxicity Prediction Based on Semi-supervised Learning and Graph Convolutional Neural Network
Summary of DeepChem Usage: DeepChem models are used to baseline the nenw method.
Important Contributions: An improvement is found in toxicity prediction accuracy.
Date Published: 2021
Journal: ResearchSquare

Paper Title: In Silico Analyses of Immune System Protein Interactome Network, Single-Cell RNA Sequencing of Human Tissues, and Artificial Neural Networks Reveal Potential …
Summary of DeepChem Usage: DeepChem is used to train fully connected networks for drug repurposingn.
Important Contributions: Repurposing is used to find potential drugs for covid-19
Date Published: 2020
Journal: Chemrxiv

Paper Title: Exploring the octanol–water partition coefficient dataset using deep learning techniques and data augmentation
Summary of DeepChem Usage: DeepChem is used heavily through the paper to develop a DNN model for predicting the octanol water coefficient.
Important Contributions: The use of DNNs for predicting the octanol-water (logP coefficient) is explored along with data augmentation and data curation for this task.
Date Published: June 2021
Journal: Nature Communications Chemistry

Paper Title: Using machine learning to classify bioactivity for 3486 per-and polyfluoroalkyl substances (PFASs) from the OECD list
Summary of DeepChem Usage: DeepChem is used for all model development work.
Important Contributions: The toxicity of per- and polyfluorinated alkyl substances is explored using machine learning methods.
Date Published: 2019
Journal: Environment Science and Technology

Paper Title: Deep learning-based prediction of drug-induced cardiotoxicity
Summary of DeepChem Usage: Datasets from MoleculeNet and DNNs/SVMs from DeepChem were used to build models.
Important Conntributions: The study develops a deep learning based approach to predict hERG blockers (deephERG) and builds a predictive model.
Date Published: April 30, 2019
Journal: J Chem Inf Modell

Paper Title: Which Hyperparameters to Optimise? An Investigation of Evolutionary Hyperparameter Optimisation in Graph Neural Network For Molecular Property Prediction
Summary of DeepChem Model Usage: DeepChem preprocessing pipelines along with DeepChem standard hyperparameters and graph conv models were used.
Important Contributions: Hyperparameter optimization techniques for graph conv models are systematically explored.
Date Published: April 14, 2021
Journal: Arxiv

Paper Title: Machine learning models for predicting endocrine disruption potential of environmental chemicals
Summary of DeepChem Usage: DeepChem featurizers were used to generate fingerprints for models.
Important Contributions: A machine learning framework for toxicity prediction is constructed for the estrogen receptor ligand bindingn domain.
Date Published: Jan 10, 2019
Journal: Journal of Environmental Science and Health

Paper Title: Dual-view Molecule Pre-training
Summary of DeepChem Usage: MoleculeNet is used to benchmark new model pretraining procedure
Important Contributions: Transformer style pretraining is combined with graph style pretraining
Date Published: June 17, 2021
Journal: Arxiv

Paper Title: Machine learning enables accurate and rapid prediction of active molecules against breast cancer cells
Summary of DeepChem Usage: DeepChem was used to train graphconv networks.
Important Contributions: Models were trained to find molecules active against breast cancer.
Date Published: September 6, 2021
Journal: Bioarxiv

Paper Title: Learning continuous and data-driven molecular descriptors by translating equivalent chemical representations
Summary of DeepChem Usage: DeepChem was used to train graphconv models.
Important Contributions: A new method is created to produce a robust molecular feature representation
Date Published: September 19th, 2018
Journal: Royal Society of Chemistry

Paper Title: Efficient lipophilicity prediction of molecules employing deep-learning models
Summary of DeepChem Usage: MoleculeNet datasets (and possibly models were used).
Important Contributions: Deep learning is used to simplify lipophilicity prediction
Date Published: June 15th, 2021
Journal: Chemometrics and Intelligent Laboratory Systems

Paper Title: FLEX: fixing flaky tests in machine learning projects by updating assertion bounds
Summary of DeepChem Usage: DeepChem unit tests are used to provide examples of flaky tests in ML software suites. This is an interesting use case of DeepChem for research that uses the raw source code and not the models/datasets themselves.
Date Published: August 23, 2021
Journal: ACM ESEC/FSE '21

Paper Title: Mol2vec: unsupervised machine learning approach with chemical intuition
Summary of DeepChem Usage: The Tox21 dataset is pulled down from MoleculeNet
Important Contributions: A new method for unsupervised learning on molecules is proposed.
Date Published: 2018
Journal: Journal of Chemical Information and Modeling

Paper Title: Improving Molecular Graph Neural Network Explainability with Orthonormalization and Induced Sparsity
Summary of DeepChem Usage: The DeepChem graphconv implementation was used as a code reference and then converted to pytorch geometric. DeepChem featurizers were used as welll.
Important Contributions: A new method for improving graphconv explainability is developed.
Date Published: August 31, 2021
Journal: Arxiv

Paper Title: In silico prediction of drug‐induced ototoxicity using machine learning and deep learning methods
Summary of DeepChem Usage: DeepChem TextCNN and another model were used to benchmark methods.
Important Contributions: A high accuracy model for drug induced ototoxicity was developed
Date Published: May 19, 2021
Journal: Chemical Biology and Drug Design

Paper Title: ACED: Accelerated Computational Electrochemical systems Discovery
Summary of DeepChem Usage: ACED is working to build a julia port of DeepChem using DeepChem weave/featurizers as references
Important Contributions: ACED is working to make a large scale automated workflow for materials discovery
Date Published: November 23, 2020
Journal: Arxiv

Paper Title: Building attention and edge message passing neural networks for bioactivity and physical–chemical property prediction
Summary of DeepChem Usage: MoleculeNet featurizers are used to generate MPNN features
Important Contributions: A new attention and edge memory scheme for MPNNs is introduced.
Date Published: January 8, 2020
Journal: Journal of Cheminformatics

Paper Title: Machine Learning Models to Predict Inhibition of the Bile Salt Export Pump
Summary of DeepChem Usage: DeepChem splitters and models were used
Important Contributions: A reliable model of drug induced liver injury is produced using the AMPL/DeepChem pipeline.
Date Published: February 28, 2020
Journal: Journal of Chemical Informatics and Modeling

Paper Title: Prediction of pharmacological activities from chemical structures with graph convolutional neural networks
Summary of DeepChem Usage: DeepChem was used extensively to featurize data and build/optimize models.
Important Contributions: Graph convolutional networks are shown to be able to predict pharmacological activity against a broad range of targets.
Date Published: January 21, 2021
Journal: Nature Scientific Reports