PSC-CPI: Multi-Scale Protein Sequence-Structure Contrasting for Efficient and Generalizable Compound-Protein Interaction Prediction (2402.08198v1)

Abstract: Compound-Protein Interaction (CPI) prediction aims to predict the pattern and strength of compound-protein interactions for rational drug discovery. Existing deep learning-based methods utilize only the single modality of protein sequences or structures and lack the co-modeling of the joint distribution of the two modalities, which may lead to significant performance drops in complex real-world scenarios due to various factors, e.g., modality missing and domain shifting. More importantly, these methods only model protein sequences and structures at a single fixed scale, neglecting more fine-grained multi-scale information, such as those embedded in key protein fragments. In this paper, we propose a novel multi-scale Protein Sequence-structure Contrasting framework for CPI prediction (PSC-CPI), which captures the dependencies between protein sequences and structures through both intra-modality and cross-modality contrasting. We further apply length-variable protein augmentation to allow contrasting to be performed at different scales, from the amino acid level to the sequence level. Finally, in order to more fairly evaluate the model generalizability, we split the test data into four settings based on whether compounds and proteins have been observed during the training stage. Extensive experiments have shown that PSC-CPI generalizes well in all four settings, particularly in the more challenging ``Unseen-Both" setting, where neither compounds nor proteins have been observed during training. Furthermore, even when encountering a situation of modality missing, i.e., inference with only single-modality protein data, PSC-CPI still exhibits comparable or even better performance than previous approaches.

• The paper introduces a novel multi-scale contrasting approach that captures protein sequence-structure dependencies for efficient CPI prediction.
• It employs intra-modality and cross-modality learning to enhance feature representation and achieve robust generalizability across unseen data.
• The framework efficiently handles both unimodal and multimodal inputs, drastically reducing computational demands in drug discovery.

Unveiling PSC-CPI: A Multi-Scale Framework for Predicting Compound-Protein Interaction through Protein Sequence-Structure Contrasting

Introduction to PSC-CPI

In the field of drug discovery, Compound-Protein Interaction (CPI) prediction remains a vital computational challenge. Conventional approaches have either relied on simulation-based methods, which are computationally intensive, or deep learning-based methods that often miss the integration of both protein sequences and structures. Addressing these limitations, the PSC-CPI (Protein Sequence-structure Contrasting for CPI prediction) framework emerges as a novel methodology. It effectively captures the dependencies between protein sequences and structures through intra-modality and cross-modality contrasting, enhancing the prediction of CPI through the innovative application of multi-scale contrasting strategies.

Key Contributions

1. Multi-Scale Contrasting: Central to PSC-CPI is its unique strategy to model protein sequence and structure dependencies at multiple scales. By applying length-variable protein augmentation, the framework contrasts information at different scales, capturing fine-grained details embedded in key protein fragments.
2. Cross-Modality Learning: The PSC-CPI framework utilizes both intra-modality and cross-modality contrasting. This approach does not only enhance the representation learning within each modality (sequence or structure) but also bridges the gap between these modalities, leveraging the benefits of multimodal information.
3. Model Generalizability: Through extensive evaluation across various dataset settings and inference situations, including those where compounds or proteins have not been seen during training, PSC-CPI demonstrates superior generalizability and robustness, outperforming traditional methods particularly in scenarios where both the compound and protein are previously unseen.
4. Efficiency in Handling Unimodal and Multimodal Data: Another significant contribution of PSC-CPI lies in its flexibility and efficiency in dealing with both unimodal (protein sequence or structure alone) and multimodal data for inference. This feature is critical for practical applications given the common scenario of modality missing in real-world datasets.

Theoretical and Practical Implications

The introduction of PSC-CPI brings forth important theoretical contributions to the field of computational biology and drug discovery. Notably, the framework’s ability to integrate and contrast multi-scale information from protein sequences and structures underlines a novel methodological pathway for CPI prediction. Furthermore, PSC-CPI's adaptability across various data splits and modalities suggests a significant advancement towards handling the inherent complexities in real-world datasets.

Practically, PSC-CPI can significantly accelerate drug discovery processes by enabling efficient and accurate prediction of CPI, particularly under challenging conditions where limited data modality is available. Such capabilities are poised to reduce the time and computational resources required for identifying potential drug candidates, thereby facilitating faster progression from computational screening to experimental validation.

Future Directions

While PSC-CPI marks a significant stride forward, it also opens avenues for further research. Exploring the application of similar contrasting strategies to other types of biomolecular interactions, extending the multi-scale modeling to finer biological details, and enhancing computational efficiency are potential areas for future work. Additionally, extending the framework to leverage unsupervised pre-training on larger unlabeled datasets could further improve its predictive performance and generalizability.

Conclusion

PSC-CPI represents a significant advance in CPI prediction, offering both theoretical insights and practical benefits for drug discovery. Through its innovative multi-scale contrasting approach and the ability to effectively utilize and integrate multimodal protein data, PSC-CPI sets a new benchmark for computational models in this domain. As the field continues to evolve, frameworks such as PSC-CPI will undoubtedly play a critical role in harnessing computational methodologies to accelerate the development of new therapeutics.