Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development (2102.09548v2)

Abstract: Therapeutics machine learning is an emerging field with incredible opportunities for innovatiaon and impact. However, advancement in this field requires formulation of meaningful learning tasks and careful curation of datasets. Here, we introduce Therapeutics Data Commons (TDC), the first unifying platform to systematically access and evaluate machine learning across the entire range of therapeutics. To date, TDC includes 66 AI-ready datasets spread across 22 learning tasks and spanning the discovery and development of safe and effective medicines. TDC also provides an ecosystem of tools and community resources, including 33 data functions and types of meaningful data splits, 23 strategies for systematic model evaluation, 17 molecule generation oracles, and 29 public leaderboards. All resources are integrated and accessible via an open Python library. We carry out extensive experiments on selected datasets, demonstrating that even the strongest algorithms fall short of solving key therapeutics challenges, including real dataset distributional shifts, multi-scale modeling of heterogeneous data, and robust generalization to novel data points. We envision that TDC can facilitate algorithmic and scientific advances and considerably accelerate machine-learning model development, validation and transition into biomedical and clinical implementation. TDC is an open-science initiative available at https://tdcommons.ai.

• The paper presents a comprehensive ML framework that integrates diverse datasets for systematic evaluation in therapeutic development.
• The study defines multiple learning tasks across target discovery, activity modeling, efficacy, safety evaluation, and manufacturing.
• The framework leverages 66 curated datasets with 15.9M data points to enable robust benchmarking using 23 evaluative metrics.

Machine Learning in Therapeutic Development: An Overview of Learning Tasks, Datasets, and Model Evaluation

The paper presents a comprehensive framework for the integration and application of ML within the lifecycle of therapeutic development. It underscores the importance of creating, curating, and leveraging datasets to formulate and evaluate ML models for diverse phases in drug discovery and development, which includes small molecules, macromolecules, and cell & gene therapies.

Key Components of the Therapeutics Lifecycle

The lifecycle of therapeutics, as detailed in this paper, is divided into various learning tasks aligned with drug discovery, activity modeling, efficacy, and safety evaluation, and manufacturing processes:

1. Target Discovery: Analyzed through five specific learning tasks, indicating the significance of identifying viable biological targets for therapeutic intervention.
2. Activity Modeling: A major focus with 13 tasks aimed at understanding biochemical interactions and stability.
3. Efficacy and Safety: Encompassing six tasks, it's crucial for assessing potential therapeutic benefits and associated risks.
4. Manufacturing: Considers four tasks to optimize the production of new therapeutics.

Data and Model Integration

The paper outlines a repository of 66 AI/ML-ready datasets derived from 15,919,332 data points, which are strategically segmented for effective ML model training and testing. The datasets include various entities such as genes, compounds, peptides, and diseases, among others. The data harmonization involves various splits tailored for different testing environments, such as random, scaffold, temporal, cold-start, and combination splits—which ensure robust model evaluation across diverse scenarios.

Model Training and Evaluation

The process of developing and testing ML models for therapeutics is aided by several sophisticated tools and evaluative functions:

• TDC Data Functions and Processing Helpers: These facilitate data preparation and transformation, crucial for high-quality model input.
• Molecule Generation Oracles and Evaluations: These orchestras help in predicting new molecular structures and benchmark them against predefined criteria, enhancing model's computational creativity and accuracy.
• TDC Evaluator Functions: A set of 23 metrics across regression, binary, multi-class, and molecular categorizations to analyze the model's performance comprehensively.

The model evaluation is benchmarked against standardized leaderboards and evaluation metrics, fostering a competitive and improvement-driven environment in ML model performance.

Implications and Future Directions

The implications of this research are multifaceted, extending both theoretical and practical advancements in AI-powered drug discovery. The structured approach to developing and validating ML models paves the way for accelerating therapeutic development. By utilizing diverse datasets and rigorous evaluation metrics, more reliable and efficient algorithms can be constructed, potentially reducing the time and cost associated with bringing new drugs to market.

Looking forward, future developments in AI within this domain may focus on refining data curation strategies, enhancing model interpretability, and integrating cutting-edge computational techniques such as reinforcement learning to further refine the drug development process. Moreover, as the volume of biomedical data continues to expand, scalable and adaptive ML models that can accommodate large, complex datasets will be imperative in maintaining the momentum in therapeutic advancements.

This paper sets a foundation for future research aiming to bridge the gap between ML and biotech advancements, providing a detailed framework that encourages collaborative innovation across interdisciplinary fields.