The field of molecular modeling and machine learning is rapidly advancing, with a focus on developing more accurate and efficient methods for predicting molecular properties and behavior. One of the key areas of research is the development of new machine learning models that can learn from large datasets of molecular structures and properties. These models have the potential to revolutionize the field of drug discovery and materials science by enabling the rapid prediction of molecular properties and the identification of new lead compounds. Notably, the use of transformers and other deep learning architectures is becoming increasingly popular in this field, allowing for the development of more accurate and generalizable models. Furthermore, researchers are exploring new methods for incorporating physical and chemical knowledge into these models, such as the use of graph neural networks and equivariant neural networks. Overall, the field of molecular modeling and machine learning is rapidly advancing, with new methods and techniques being developed continuously. Noteworthy papers in this area include HIP, which demonstrates the ability to predict Hessians directly from a deep learning model, and MolSpectLLM, which achieves state-of-the-art performance on spectrum-related tasks by explicitly modeling molecular spectra. Additionally, GRAM-TDI and MCGM propose innovative approaches to drug target interaction prediction and long-range interaction modeling, respectively.
Advances in Molecular Modeling and Machine Learning
Sources
Shoot from the HIP: Hessian Interatomic Potentials without derivatives
MolSpectLLM: A Molecular Foundation Model Bridging Spectroscopy, Molecule Elucidation, and 3D Structure Generation
GRAM-TDI: adaptive multimodal representation learning for drug target interaction prediction
MCGM: Multi-stage Clustered Global Modeling for Long-range Interactions in Molecules
SoDaDE: Solvent Data-Driven Embeddings with Small Transformer Models
Learning the Neighborhood: Contrast-Free Multimodal Self-Supervised Molecular Graph Pretraining
Toward a Robust Biomimetic Hybrid Battery: Bridging Biology, Electrochemistry and Data-Driven Control
Revisit the Imbalance Optimization in Multi-task Learning: An Experimental Analysis
DiBS-MTL: Transformation-Invariant Multitask Learning with Direction Oracles
ADAPT: Lightweight, Long-Range Machine Learning Force Fields Without Graphs
Is Sequence Information All You Need for Bayesian Optimization of Antibodies?
SOLD: SELFIES-based Objective-driven Latent Diffusion
MSCoD: An Enhanced Bayesian Updating Framework with Multi-Scale Information Bottleneck and Cooperative Attention for Structure-Based Drug Design
Can Molecular Foundation Models Know What They Don't Know? A Simple Remedy with Preference Optimization
AIM: Adaptive Intervention for Deep Multi-task Learning of Molecular Properties
MC-GNNAS-Dock: Multi-criteria GNN-based Algorithm Selection for Molecular Docking
Learning Inter-Atomic Potentials without Explicit Equivariance
Neural Network Surrogates for Free Energy Computation of Complex Chemical Systems
Round-trip Reinforcement Learning: Self-Consistent Training for Better Chemical LLMs
Catalyst GFlowNet for electrocatalyst design: A hydrogen evolution reaction case study
Transformers Discover Molecular Structure Without Graph Priors