Hey all,
How can i use the model for enhancer-gene predictions? i read the paper and i know you have done extensive benchmarking on this, and i know how to make RNA-seq predictions, but i just can’t find a clean way to do enhancer-gene prediction. it seems like there is no function for it. can you help me with that please?
Thank you for your time, Mike
Hi Mike,
Thank you for reaching out.
There is currently no dedicated function for enhancer-gene prediction via the API. The primary method described in the paper utilized input gradients (saliency maps) to score links. The AlphaGenome code has now been published and the pre-trained model weights are available. You can now run inference locally by defining a large input window (eg 1Mb) around a gene, predicting gene-associated outputs (such as RNA-seq), and then computing input x gradient with respect to the predicted output. This produces a base-resolution influence map across the input interval, which can be aggregated over candidate regulatory regions to score enhancer-gene links.
Kind regards,
Tumi