Molecular Machine Learning

Artificial intelligence (AI), particularly machine learning (ML), holds great promise in revolutionizing molecule and material design by accelerating the discovery of new compounds with desired properties. Our lab is actively involved in the following research topics in molecular/material ML:

Property prediction: Accurate and efficient prediction of various molecular and material properties is essential for accelerating the design of new compounds. However, obtaining high-quality labeled data for training ML models can be challenging due to the expense and time-consuming simulations and experiments. To address this, our lab is developing self-supervised learning (SSL) methods that leverage large amounts of unlabeled data to build better ML models, enhancing property prediction accuracies.
Molecular simulations: AI can be applied to molecular simulations to accelerate molecular simulations. Our lab is working on building cost-effective ML surrogate models for molecular force fields with quantum mechanics level accuracy.
Optimization of molecules/materials: Our lab is utilizing deep reinforcement learning and deep generative models to optimize molecules and materials towards desired properties. ML property predictors are used as surrogate models in an in-loop framework, allowing for efficient and effective optimization of compounds.
De novo generation: Deep generative models are employed for de novo design of molecules and materials, which enable exploration of new chemical space. Our lab is leveraging these models to generate novel compounds with desired properties, leading to the discovery of new materials and molecules.
Model Interpretability and Explainability: Deep neural networks are often considered as "black boxes" due to their complex nature. Our lab also investigate building interpretable AI models that incorporate domain knowledge and provide insights to researchers, enhancing the interpretability and explainability of ML models in molecular and material design.

By leveraging AI and ML techniques in these areas, our lab aims to accelerate the design and discovery of molecules and materials with desired properties, overcome data limitations, improve simulation accuracy and efficiency, optimize compounds, generate novel compounds, and enhance model interpretability and explainability.

Our research at MAIL

Chemical Language Model

We leverage the power of Transformer architecture to generate, analyze, and predict chemical compounds and material properties:

Xu, C., Wang, Y., & Farimani, A. B. (2023). TransPolymer: a Transformer-based Language Model for Polymer Property Predictions. npj Computational Materials.
Cao, Z., Magar, R., Wang, Y., & Barati Farimani, A. (2023). MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction. Journal of the American Chemical Society, 145(5), 2958-2967.

Self-Supervised Learning

Self-supervised learning (SSL) addresses the challenge of limited labeled data in chemistry and materials science, allowing for improved accuracy and generalization of ML models. We develop various SSL models for molecule/material science.

Wang, Y., Xu, C., Li, Z., & Farimani, A. B. (2023). Denoise Pre-training on Non-equilibrium Molecules for Accurate and Transferable Neural Potentials. arXiv preprint arXiv:2303.02216.
Magar, R., Wang, Y., & Barati Farimani, A. (2022). Crystal twins: self-supervised learning for crystalline material property prediction. npj Computational Materials, 8(1), 231.
Wang, Y., Magar, R., Liang, C., & Barati Farimani, A. (2022). Improving molecular contrastive learning via faulty negative mitigation and decomposed fragment contrast. Journal of Chemical Information and Modeling, 62(11), 2713-2725.
Wang, Y., Wang, J., Cao, Z., & Barati Farimani, A. (2022). Molecular contrastive learning of representations via graph neural networks. Nature Machine Intelligence, 4(3), 279-287.

AI in Material Design

We develop deep reinforcement learning (DRL) agents to learn from interactions with the environment, such as simulating material structures and properties, and iteratively optimize material designs based on simulated feedbacks:

Yoon, J., Cao, Z., Raju, R. K., Wang, Y., Burnley, R., Gellman, A. J., Barati Farimani, A., & Ulissi, Z. W. (2021). Deep reinforcement learning for predicting kinetic pathways to surface reconstruction in a ternary alloy. Machine Learning: Science and Technology, 2(4), 045018.
Wang, Y., Cao, Z., & Barati Farimani, A. (2021). Efficient water desalination with graphene nanopores obtained using artificial intelligence. npj 2D Materials and Applications, 5(1), 66.

Property prediction

We introduce domain knowledge and interpretable mechanism to “black-box” neural networks for the purpose of building more accurate models and providing insights to researchers:

Jian, Y., Wang, Y., & Barati Farimani, A. (2022). Predicting CO2 Absorption in Ionic Liquids with Molecular Descriptors and Explainable Graph Neural Networks. ACS Sustainable Chemistry & Engineering, 10(50), 16681-16691.
Magar, R., Wang, Y., Lorsung, C., Liang, C., Ramasubramanian, H., Li, P., & Farimani, A. B. (2022). AugLiChem: data augmentation library of chemical structures for machine learning. Machine Learning: Science and Technology, 3(4), 045015.
Karamad, M., Magar, R., Shi, Y., Siahrostami, S., Gates, I. D., & Farimani, A. B. (2020). Orbital graph convolutional neural network for material property prediction. Physical Review Materials, 4(9), 093801.

Molecular Simulation

Our lab works on molecular dynamics simulations to understand different molecular systems. Also, we are investigating acceleration of molecular simulations with AI/ML models:

Li, Z., Meidani, K., Yadav, P., & Barati Farimani, A. (2022). Graph neural networks accelerated molecular dynamics. The Journal of Chemical Physics, 156(14), 144103.
Mei, L., Cao, Z., Ying, T., Yang, R., Peng, H., Wang, G., Zheng, L., Chen, Y., Tang, CY., Voiry, D., Wang, H., Barati Farimani, A., & Zeng, Z. (2022). Simultaneous electrochemical exfoliation and covalent functionalization of MoS2 membrane for ion sieving. Advanced Materials, 34(26), 2201416.
Cao, Z., Yadav, P., & Barati Farimani, A. (2022). Which 2D Material is Better for DNA Detection: Graphene, MoS2, or MXene?. Nano Letters, 22(19), 7874-7881.
Meidani, K., Cao, Z., & Barati Farimani, A. (2021). Titanium carbide MXene for water desalination: a molecular dynamics study. ACS Applied Nano Materials, 4(6), 6145-6151.
Cao, Z., Liu, V., & Barati Farimani, A. (2020). Why is single-layer MoS2 a more energy efficient membrane for water desalination?. ACS Energy Letters, 5(7), 2217-2222.
Cao, Z., Liu, V., & Barati Farimani, A. (2019). Water desalination with two-dimensional metal–organic framework membranes. Nano letters, 19(12), 8638-8643.

To know more about AI and Material Discovery, you can watch these videos

Google Sites

Report abuse