miRe2e is a novel deep learning model based on Transformers that allows finding pre-miRNA sequences in raw genome-wide data. This model is a full end-to-end neural architecture, using only the raw sequences as inputs. This demo provides a miRe2e model pre-trained with known pre-miRNAs from H. sapiens, capable of classifying candidate sequences to be novel pre-miRNAs.
Input FASTA files can be up to 1 kbp for this demo.The raw sequence is analyzed in short segments, and a score for each segment is calculated.
Each sequence is analyzed in both directions: 5’ to 3’ and 3’ to 5’. For each direction, each position in the sequence gets a score. The higher the score, the higher the probability of a pre-miRNA sequence.
Sample inputs from the 19 chromosome:
- Sample 1: sequence with pre-miRNAs hsa-mir-24-2 and hsa-mir-27a in the positions 13,836,287 and 13,836,440 of strand 3′ respectively.
- Sample 2: sequence without pre-miRNAs.
For using larger sequences, please follow the instructions provided with the source code.
Contact: Jonathan Raad