GAME: Genomic API for Model Evaluation

Abstract

The rapid expansion of genomics datasets and the application of machine learning has produced sequence-to-activity genomics models with ever-expanding capabilities. However, benchmarking these models on practical applications has been challenging because individual projects evaluate their models in ad hoc ways, and there is substantial heterogeneity of both model architectures and benchmarking tasks. To address this challenge, we have created GAME, a system for large-scale, community-led standardized model benchmarking on user-defined evaluation tasks. We borrow concepts from the Application Programming Interface (API) paradigm to allow for seamless communication between pre-trained models and benchmarking tasks, ensuring consistent evaluation protocols. Because all models and benchmarks are inherently compatible in this framework, the continual addition of new models and new benchmarks is easy. We also developed a Matcher module powered by a large language model (LLM) to automate ambiguous task alignment between benchmarks and models. Containerization of these modules enhances reproducibility and facilitates the deployment of models and benchmarks across computing platforms. By focusing on predicting underlying biochemical phenomena (e.g. gene expression, open chromatin, DNA binding), we ensure that tasks remain technology-independent. We provide examples of benchmarks and models implementing this framework, and anticipate that the community will contribute their own, leading to an ever-expanding and evolving set of models and evaluation tasks. This resource will accelerate genomics research by illuminating the best models for a given task, motivating novel functional genomic benchmarks, and providing a more nuanced understanding of model abilities.

Figure 1:. GAME framework.

GAME includes three modules: The Evaluator, containing a benchmark dataset; the Predictor, encompassing a sequence-to-activity model; and the Matcher, capturing relationships between tasks. All GAME modules are inherently interoperable by communicating in the GAME API protocol over TCP. For each benchmark, the Evaluator requests a prediction from the Predictor, which consults the Matcher to determine the closest task the Predictor can complete. Once the Matcher returns the best match, the Predictor will complete its prediction and return it to the Evaluator, which will evaluate performance. Members of the genomics community will contribute modules to enable continual evaluation of more models across more benchmarks.

Figure 2:. Sample benchmarking done with GAME.

a, Expression evaluation tasks. Models (x axis) were evaluated for their correlation to measured expression levels (colours) across a variety of tasks (y axis). b, Chromatin conformation tasks. Correlation of Orca predictions vs measured chromatin contact frequencies (colours) for two Orca test-set chromosomes and one validation-set chromosome (y axis). c, Consistency evaluation for accessibility. Point and track based accessibility consistency evaluators (y axis) were used to evaluate the correlation between predictions for forward and reverse complement sequences (colours) in models of DNA accessibility (x axis). *Correlation values from down sampled datasets. Grey cells mark values that were not calculated due models that could not complete the Evaluator’s requested task.