DeepRC

DeepRC classifies immune repertoires using transformer-like attention mechanisms and modern Hopfield networks within deep learning architectures to address massive multiple instance learning (MIL) problems in immunosequence analysis.

Key Features:

• Transformer-like Attention Mechanism: Utilizes an attention mechanism akin to transformer architectures that functions as the update rule of modern Hopfield networks, enabling storage and retrieval of exponentially many patterns.
• Modern Hopfield Networks: Incorporates modern Hopfield networks to capitalize on high pattern storage capacity for processing large numbers of immunosequences.
• Deep Learning Architectures for Massive MIL: Employs deep learning architectures specifically tailored for massive multiple instance learning tasks.
• Massive MIL Capability: Engineered to handle unprecedentedly large numbers of instances with very low witness rates typical of immune repertoire classification.
• Predictive Performance: Demonstrated superior predictive performance in large-scale experiments on simulated and real-world virus infection data, outperforming existing methods.
• Interpretability and Motif Extraction: Facilitates extraction of sequence motifs linked to specific disease classes to support biological interpretation.

Scientific Applications:

• Immune Repertoire Classification: Classification of immune repertoires from individual immunosequences using MIL approaches.
• Vaccine and Therapy Research: Supports research into vaccines and therapies by enabling detailed analysis of immune system responses.
• Virus Infection Data Analysis: Applied to simulated and real-world virus infection datasets for predictive analysis of infection-associated repertoires.
• Motif Discovery: Identification of sequence motifs associated with disease classes to inform mechanistic studies of immune responses.

Methodology:

Integrates transformer-like attention mechanisms (functionally equivalent to modern Hopfield networks) into deep learning architectures to enable pattern storage and retrieval and to manage and analyze massive MIL datasets typical of immune repertoire classification.