Gast ::
| Hauptseite > Publikationsdatenbank > Machine Learning Guided RNA Structure Prediction |
| Hauptseite > Publikationsdatenbank > Machine Learning Guided RNA Structure Prediction |
| Poster (After Call) | FZJ-2023-05499 |
; ; ;
2023
Abstract: For around 50 years, the primary focus of genomic research has been the development ofefficient and accurate methods to predict the structure of proteins, which led to the birth ofbetter sequencing techniques and databases. About 98% of the human genome (RNA, DNA)during this action was overlooked.Many consider RNA merely as a messenger between DNA and ribosomes for making proteins.However, In the past few years, studies have revealed the existence of many non-coding RNAswhich catalyse various biological processes; to gain detailed insights into these roles, werequire the appropriate structure of RNAs. Recent years have led to breakthroughs in proteinstructure prediction via Deep Learning. The scarcity of RNA structures, however, makes adirect transfer of these methods impossible. Here, we present machine-learning techniquesthat can work with limited training data. We predict contact maps as a proxy to understandand predict RNA structure, they provide a minimal representation of the structure. We haveworked on methods that took accuracy from 47%(DCA) to 77%(CoCoNet) and now to87%(Barnacle) i.e. doubling accuracy while reducing false positives by five-fold. Further,research is going on to create much more efficient neural networks which make use ofstatistical physics and ML techniques like Attention mechanisms and Transformers. We areconfident that this remarkable progress will reduce the sequence-structure gap for RNA.
|
The record appears in these collections: |