de Boer Lab @ UBC

GAME was designed for the functional genomics community to create seamless communication across pre-trained models and genomics datasets. It is a product of the feedback from many model and dataset experts and our hope is that it allows for long-lasting benchmarking of models.

Trained on human RNA-seq data from ENCODE (with 866 datasets across diverse biosamples, including cell lines and adult tissues) and GTEx data (with 2-3 replicates for each tissue, processed by the recount3 project). The training dataset also includes epigenomic datasets from Enformer, such as CAGE, DNase-seq, ATAC-seq, and ChIP-seq tracks.

To evaluate whether models were internally consistent in their predictions, we created Consistency Evaluators.

Variant Position Modulates Functional Impact in Massively Parallel Reporter Assays

The Enformer module encapsulates the human and mouse head of the large-scale multi-task Enformer model and can make predictions for sequences of any length. Its training data includes chromatin accessibility, DNA binding and expression.

Agarwal et al. (2025) MPRA evaluator first requests log-scale predictions for ~56k probes across WTC11, K562, and HepG2 cells. It then calculates Pearson correlations between predicted and measured (log2(rna/dna)). It also assesses cell-type specificity by correlating predicted vs. observed expression differences.

Martyn et al. dataset measured the effects of 783 variants in Jurkat T cells and THP-1 monocytes cells. This Evaluator is used to evaluate models' abilities to predict variant effects. We extract 2000bp around the indel and request log scale expression predictions for the reference and mutated sequence.

As most of the sample Evaluators request point-based predictions, we incorporated an Evaluator that explicitly requests and evaluates track-based predictions. To construct this Evaluator, we intersected and merged genomics coordinates from the Enformer and Borzoi test datasets.