De novo molecule design towards biased properties via a deep generative framework and iterative transfer learning

As quantum mechanics marks its centennial in 2025, machine learning interatomic potentials have emerged as transformative tools in molecular modeling, bridging quantum mechanical accuracy with classical efficiency. Here we examine their development through four defining challenges-achieving chemical accuracy, maintaining computational efficiency, ensuring interpretability and reaching universal generalizability. We highlight architectural innovations, physics-informed approaches, and foundation models trained on extensive data. Together, these developments chart a path toward predictive, transferable and physically grounded machine learning frameworks for next-generation computational chemistry.

Quantitative structure-activity relationship (QSAR) modelling, an approach that was introduced 60 years ago, is widely used in computer-aided drug design. In recent years, progress in artificial intelligence techniques, such as deep learning, the rapid growth of databases of molecules for virtual screening and dramatic improvements in computational power have supported the emergence of a new field of QSAR applications that we term 'deep QSAR'. Marking a decade from the pioneering applications of deep QSAR to tasks involved in small-molecule drug discovery, we herein describe key advances in the field, including deep generative and reinforcement learning approaches in molecular design, deep learning models for synthetic planning and the application of deep QSAR models in structure-based virtual screening. We also reflect on the emergence of quantum computing, which promises to further accelerate deep QSAR applications and the need for open-source and democratized resources to support computer-aided drug design.

There has been a recent focus within Alzheimer's Disease (AD) research on potential CSF biomarkers as diagnostic tools. PET imaging and CSF biomarkers have been used to determine the amyloid status of individuals when characterising AD. While these measures have largely correlated with each other, it is also suggested that there could be patients who can have discordant results. However, it is still unclear what proportion of AD subjects demonstrate the discordance. Additionally, whether these changes are associated with different pathological and clinical characteristics of the patients. Here, we investigated the discordance between CSF Aβ42:40 and amyloid status and evaluated whether there are any significant changes in their levels of tau aggregation. 313 participants' data were selected from the ADNI database. The cutoff for Aβ42:40 was then calculated using the Youden Index resulting from receiver operating characteristic (ROC) curve. The formula used was J = maxc Se (c) + Sp (c) - 1, resulting in a maximum cutoff of J = 0.057 for Aβ42:40. This cutoff produced a discordance rate of 10.5%, with n = 33 of 313 patients being discordant. We then performed t-tests to investigate the potential clinical differences between concordant and discordant patients. Independent samples t-tests indicated that levels of pTau181 and total Tau significantly differed between the concordant and discordant groups. pTau181 was significantly lower in the discordant group (t(71.2)=-3.166, p = .002), as was Tau (t(53.9)=-2.046, p = .046). Further data is found in Figure 1. Demonstration of different levels of tau deposition in the discordant groups implies there may be other pathological processes influencing neurodegeneration in these subjects. A discordance of 10% between CSF amyloid and PET amyloid implies that we should evaluate these subjects in greater detail before enrolling them in intervention studies. References 1. Hansson, O. et al. (2019). https://doi.org/10.1186/s13195-019-0485-0 2. Pyun, JM. et al. (2024) https://doi.org/10.1038/s41398-024-02766-6.

ConspectusMachine learning interatomic potentials (MLIPs) are widely used for describing molecular energy and continue bridging the speed and accuracy gap between quantum mechanical (QM) and classical approaches like force fields. In this Account, we focus on the out-of-the-box approaches to developing transferable MLIPs for diverse chemical tasks. First, we introduce the "Accurate Neural Network engine for Molecular Energies," ANAKIN-ME, method (or ANI for short). The ANI model utilizes Justin Smith Symmetry Functions (JSSFs) and realizes training for vast data sets. The training data set of several orders of magnitude larger than before has become the key factor of the knowledge transferability and flexibility of MLIPs. As the quantity, quality, and types of interactions included in the training data set will dictate the accuracy of MLIPs, the task of proper data selection and model training could be assisted with advanced methods like active learning (AL), transfer learning (TL), and multitask learning (MTL).Next, we describe the AIMNet "Atoms-in-Molecules Network" that was inspired by the quantum theory of atoms in molecules. The AIMNet architecture lifts multiple limitations in MLIPs. It encodes long-range interactions and learnable representations of chemical elements. We also discuss the AIMNet-ME model that expands the applicability domain of AIMNet from neutral molecules toward open-shell systems. The AIMNet-ME encompasses a dependence of the potential on molecular charge and spin. It brings ML and physical models one step closer, ensuring the correct molecular energy behavior over the total molecular charge.We finally describe perhaps the simplest possible physics-aware model, which combines ML and the extended Hückel method. In ML-EHM, "Hierarchically Interacting Particle Neural Network," HIP-NN generates the set of a molecule- and environment-dependent Hamiltonian elements αμμ and K‡. As a test example, we show how in contrast to traditional Hückel theory, ML-EHM correctly describes orbital crossing with bond rotations. Hence it learns the underlying physics, highlighting that the inclusion of proper physical constraints and symmetries could significantly improve ML model generalization.