We propose an overfitting-based hypothesis for the latter finding, and study it empirically. Through benchmarks on diverse tasks, we find post-hoc conjoined consistently performs best or second-best, surpassed only occasionally by RCPS, and never underperforms conjoined-during-training. In this work, we extend conjoined & RCPS models to base-resolution signal prediction, and introduce a strong baseline: a standard model (trained with RC data augmentation) that is made conjoined only after training, which we call “post-hoc” conjoined. However, the connections between the two remain unclear, comparisons to strong baselines are lacking, and neither has been adapted to base-resolution signal profile prediction. ![]() Two strategies have emerged to enforce equivariance: conjoined/“siamese” architectures, and RC parameter sharing or RCPS. Unfortunately, standard neural networks can produce highly divergent predictions across strands, even when the training set is augmented with RC sequences. %X Predictive models mapping double-stranded DNA to signals of regulatory activity should, in principle, produce analogous (or “equivariant”) predictions whether the forward strand or its reverse complement (RC) is supplied as input. %C Proceedings of Machine Learning Research %B Proceedings of the 16th Machine Learning in Computational Biology meeting %T Towards a Better Understanding of Reverse-Complement Equivariance for Deep Learning Models in Genomics A 22-minute video explaining the paper is available at Ĭite this = The code to replicate the experiments is available at. Finally, we present a unified description of conjoined & RCPS architectures, revealing a broader class of models that gradually interpolate between RCPS and conjoined while maintaining equivariance. Our results suggest users interested in RC equivariance should default to post-hoc conjoined as a reliable baseline before exploring RCPS. Despite its theoretical appeal, RCPS shows mediocre performance on several tasks, even though (as we prove) it can represent any solution learned by conjoined models. It can be redistributed under the same terms as EMBOSS itself.Predictive models mapping double-stranded DNA to signals of regulatory activity should, in principle, produce analogous (or “equivariant”) predictions whether the forward strand or its reverse complement (RC) is supplied as input. ![]() This manual page was autogenerated from an Ajax Control Definition of the EMBOSS package. Wrote the script used to autogenerate this manual page. Revseq is fully documented via the tfm(1) system. Set this to be false if you do not wish to complement the output sequence Defaultīugs can be reported to the Debian Bug Tracking system ( ), or Set this to be false if you do not wish to reverse the output sequence Default value: OPTIONS Input section -sequence seqall Advanced section -reverse boolean It is part of the "Edit" command group(s). SYNOPSIS revseq -sequence seqall -reverse boolean -complement boolean -outseq seqoutall revseq -help DESCRIPTION revseq is a command line program from EMBOSS (“the European Molecular Biology Open Revseq - Reverse and complement a nucleotide sequence
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |