Computational Advances in Structural Biology

Abstract

Recent developments in artificial intelligence (AI) have opened up new avenues for accelerating drug discovery. Deep learning techniques have demonstrated immense potential in addressing significant issues, such as affinity prediction, conformer generation, and pose prediction. These techniques help to identify potential drug candidates faster and more effectively than traditional methods. However, the efficacy of deep learning algorithms is frequently restricted by the availability of high-quality data for training. Despite attempts to generate more data, the limited number of examples, experimental errors, and the lack of variety in available data can create biases that may influence the performance of deep learning algorithms in real-world situations.

This presentation will emphasize the importance of tackling data restrictions and biases in deep learning techniques, using affinity prediction and docking algorithms as examples. Although machine learning techniques have produced remarkable outcomes in benchmark studies, our results suggest that they may not perform as well in real-world applications. I will also explore how experimental structural information from Fragment-Based Drug Discovery can be leveraged to overcome these data limitations and enhance the performance of deep learning algorithms. Finally, I will present some case studies that demonstrate the potential of this approach for accelerating drug discovery.