Molecular Property Prediction Survey Overview

Paper context

Paper title: A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the Foundation Model Era Abstract: Molecular property prediction integrates quantum chemistry, cheminformatics, and deep learning to connect molecular structure with physicochemical and biological behavior. This survey traces four complementary paradigms, including Quantum, Descriptor Machine Learning, Geometric Deep Learning, and Foundation Models, and outlines a unified taxonomy linking molecular representations, model architectures, and interdisciplinary applications. Benchmark analyses integrate evidence from both widely used datasets and datasets reflecting industry perspectives, encompassing quantum, physicochemical, physiological, and biophysical domains. The survey examines current standards in data curation, splitting strategies, and evaluation protocols, highlighting challenges including inconsistent stereochemistry, heterogeneous assay sources, and reproducibility limitations under random or poorly defined splits. These observations motivate the modernization of benchmark design toward more transparent, time- and scaffold-aware methodologies. We further propose three forward-looking directions: (i) physics-aware learning embedding quantum consistency, (ii) uncertainty-calibrated foundation models for trustworthy inference, and (iii) realistic multimodal benchmark ecosystems integrating computational and experimental data. Repository: https://github.com/Zongru-Li/Survey-and-Benchmarks-of-DL-for-Molecular-Property-Prediction-in-the-Foundation-Model-Era. Passages referencing this figure: oss multiple datasets, emphasizing the impact of different data splitting strategies on model performance. Section 7 discusses practical applications across major chemical and biological domains. Finally, Section 8 outlines the roadmap and identifies promising future research directions. The complete framework, encompassing evolution, taxonomy, benchmark evaluation, and roadmap, is illustrated in Figure 1 . 2 Preliminaries and Problem Definition 2.1 Molecular Property Prediction Task Definition Molecular property prediction (MPP) refers to learning a mapping from a molecular representation to one or more properties or activities of interest ( 74 ) . Formally, let m ∈ ℳ m\in\mathcal{M} denote an input representation of a molecule (e.g., a graph, sequence, or geometric point cloud) and let y or ecosystems, as well as their biodegradability and environmental persistence ( 119 ) . Across these diverse application domains, the prediction problem is fundamentally framed as establishing structure-property relationships. Given a molecular structure represented in an appropriate format, the task is to predict an associated label or numerical value corresponding to the property of interest. Figure 1: Overview of this survey. The framework organizes more than 100 deep learning methods for molecular property prediction along four axes: Evolution , Taxonomy , Capability , and Roadmap . From left to right: (i) Evolution traces the cumulative methodological trajectory from quantum mechanics and descriptor-based learning to geometric and foundation models; (ii) Taxonomy categorizes methods during model development and validation ( 6 , 26 ) . These challenges highlight the need for benchmarks that better reflect real-world complexity and support the development of uncertainty-aware, exp