Aug 2026· IEEE transactions on computational biology and bioinformatics· Vol PP, pp. 1-13· 0 citations
Medicine
TL;DR
MolPACL is proposed, a prompt-augmentation-based supervised contrastive learning framework that incorporates high-level chemical semantics while preserving molecular identity, and achieves strong performance on both classification and regression tasks while reducing training cost.
Abstract
Large language models (LLMs) provide a promising direction for learning molecular representations from text-like inputs, yet most molecular contrastive learning methods still rely on graph- or SMILES-level augmentations that may unintentionally distort chemical structure. We propose MolPACL, a prompt-augmentation-based supervised contrastive learning framework that incorporates high-level chemical semantics while preserving molecular identity. MolPACL generates multiple semantically consistent prompt views for each molecule from its SMILES string and physicochemical descriptors using diverse templates and lightweight lexical perturbations. These views are combined with task-aware class positives and negatives to form contrastive batches, and the model is trained using a supervised objective based on the Soft Nearest Neighbor loss. Experiments on MoleculeNet benchmarks show that the proposed approach achieves strong performance on both classification and regression tasks while reducing training cost, requiring no additional molecular pretraining and using a relatively small pretrained LLM.
Transformer models have revolutionized natural language processing (NLP), and text-based molecular representations like SMILES have successfully extended these architectures to chemistry. However, domain-adaptive pre-training often causes models to overfit to chemical syntax, catastrophically forgetting their foundational semantic capabilities. To address this challenge, we introduce CheMatE, a chemistry-oriented embedding model that jointly captures molecular structure and domain-specific natural language within the same representation space. Built on a ModernBERT backbone, CheMatE learns bi-semantic representations through a two-stage training procedure: continued masked language modeling (MLM) followed by a Matryoshka contrastive learning stage via Multiple Negative Ranking Loss (MNRL). First, we train the model using MLM on a novel, large-scale corpus of SMILES-annotated, long-context scientific documents that were constructed and curated from FineWeb and ChemPile (comprising 10.4B and 11.5B tokens, respectively). Subsequently, the model undergoes contrastive learning using a synthetic dataset of SMILES-text pairs algorithmically derived from our original training corpus. This design exposes the model to SMILES-enriched scientific literature, enabling bi-semantic understanding. We evaluate CheMatE across a range of downstream tasks covering molecular property prediction and scientific language understanding. Our results demonstrate that coupling our custom-curated datasets with this sequential training strategy yields robust, highly transferable representations. By effectively unifying structural and contextual signals within a single text-based framework, CheMatE achieves competitive performance across both specialized chemistry models and general-purpose language model baselines.
David Ming Segura, Jeremy Goumaz, Joshua W. Sin et al.· 0 citations
Although existing molecular pretraining models have achieved favorable performance on various downstream tasks, their reliance on explicit three-dimensional conformer sampling or external natural-language corpora often increases computational cost and affects structural fidelity. Here, we propose ClipMol, a molecular representation learning framework based on SMILES-InChI dual-view chemical-language alignment. Without requiring explicit three-dimensional conformers or external corpora, ClipMol jointly models local chemical microenvironments and global structural constraints of molecules. Benchmark results show that ClipMol and its scaled variant, ClipMol-XL, achieve strong overall performance on both classification and regression tasks. More importantly, for collision cross-section (CCS) prediction in ion mobility-mass spectrometry, ClipMol shows stable and competitive performance on two independent benchmark data sets, METLIN-CCS and ALLCCS, while maintaining robustness across different adduct compositions and diverse chemical categories. Compared with state-of-the-art and competitive CCS prediction models, the ClipMol models achieved the best or highly competitive averaged performance on both data sets, with ClipMol-XL showing the strongest overall R2 and root-mean-square error performance. Overall, ClipMol provides a scalable and structurally faithful solution for molecular representation learning and IM-MS-related CCS prediction in analytical chemistry.
Shengcai Shen, Chaoting Shi, Yu Wang et al.· Analytical Chemistry· 0 citations
This paper proposes TextDTI, a multimodal framework that simultaneously exploits sequential and structural representations and enhances feature alignment through adversarial learning and contrastive loss, resulting in robust and high-performance DTI prediction.
Jiaqi Deng, Senyu Tang, Jijun Tang et al.· Journal of Chemical Informat...· 0 citations
Molecular deep learning plays an important role in addressing challenging molecular property prediction tasks. However, labeled molecular data remain scarce, and the majority of existing studies predominantly employ single-modal methods. Most single-modal models face limitations in simultaneously capturing molecular topological features and modeling long-range dependencies in sequences. In this study, we propose a novel multimodal alignment framework for joint modeling of molecular graphs and sequences, called Mol-ME. The framework incorporates a data augmentation strategy to enhance model performance under limited labeling conditions. Mol-ME comprises four core modules. The first module consists of dual encoders that generate graph-based and sequence-based molecular representations, which are then aligned through contrastive learning. The second module, a gated cross-modal fusion network, enables fine-grained integration of these representations by leveraging both the cross-attention mechanism and the gating mechanism. The third module is a motif-aware feature extractor that captures latent relationships among molecular substructures. The final module employs ensemble learning to predict on extracted representations, which captures complex nonlinear relationships and compensates for the modeling limitations of single shallow networks. Experimental results on 9 benchmark data sets demonstrate that Mol-ME consistently outperforms all baseline methods, achieving new state-of-the-art (SOTA) performance in molecular property prediction.
Baoren Huang, Mu Chen, Junjie Luo et al.· Journal of Chemical Informat...· 0 citations
This is the first method to expose GNN-derived attributions to an LLM as evidence for property prediction, and achieves the best overall results among generalist models and narrows the gap to specialist models tuned for each task.
Junwoo Park, Minyoung Shin, C. Lee et al.· 0 citations
Molecular property prediction is central to cheminformatics and environmental chemistry, where accurate modeling of physicochemical properties supports risk assessment and molecular design. Classical descriptors and recent advances such as ChemBERTa have enabled learning chemically contextual representations directly from SMILES, while the integration of structured descriptors with transformer-based embeddings offers a promising pathway toward accurate and interpretable prediction. In this study, we introduce Path-Weighted Atom Vectors (PWAVs), a descriptor family that captures atom-level, environment-aware structural information. We evaluate PWAV both as a standalone representation and in combination with ChemBERTa embeddings through a gated fusion architecture incorporating modality dropout, FiLM conditioning, and auxiliary supervision. Experiments on six physicochemical property datasets ( log P, log S, log BCF, boiling point, melting point, and vapor pressure) show that PWAV generally improves over classical fingerprint descriptors within learned models and achieves competitive performance relative to established external baselines on several endpoints. The strongest gains are observed for boiling point, aqueous solubility, and partition coefficient prediction, where descriptor-embedding fusion yields the best results among the learned models considered. Ablation analyses demonstrate that PWAV contributes complementary structural information beyond SMILES-only ChemBERTa representations, while SHapley Additive exPlanations-based interpretability shows that predictive signal is concentrated within a compact subset of features, enabling an efficient reduced representation (PWAV-64). Nested cross-validation further confirms the robustness of PWAV within the XGBoost framework. Overall, PWAV provides a compact, interpretable, and extensible descriptor framework that integrates effectively with modern representation-learning approaches. These results position PWAV as a competitive and chemically transparent component for hybrid molecular property prediction, rather than as a replacement for domain-specific benchmark systems.
M. Afzal, S. Siddiqi· Physica Scripta· 0 citations