A deep-learning system combining transformers and convolutional neural networks accelerated drug-target affinity prediction and virtual screening for repurposing candidates. The approach aims to narrow down binding hypotheses computationally, reducing the experimental burden for finding actionable targets. Separately, an AI blood-based model (Fragmentia-AI-COAD) was described as detecting colorectal cancer and advanced adenomas using cell-free DNA fragmentomic signatures. The clinical promise here is earlier, less invasive detection compared with colonoscopy-based pathways. Together, the two items reinforce how AI models are being operationalized across the discovery-to-diagnostics pipeline—affinity prediction supports early candidate ranking, while fragmentomic patterning supports minimally invasive disease detection.