Mutate and observe: utilizing deep neural networks to investigate the impact of mutations on translation initiation

Abstract

Motivation The primary regulatory step for protein synthesis is translation initiation, which makes it one of the fundamental steps in the central dogma of molecular biology. In recent years, a number of approaches relying on deep neural networks (DNNs) have demonstrated superb results for predicting translation initiation sites. These state-of-the art results indicate that DNNs are indeed capable of learning complex features that are relevant to the process of translation. Unfortunately, most of those research efforts that employ DNNs only provide shallow insights into the decision-making processes of the trained models and lack highly sought-after novel biologically relevant observations. Results By improving upon the state-of-the-art DNNs and large-scale human genomic datasets in the area of translation initiation, we propose an innovative computational methodology to get neural networks to explain what was learned from data. Our methodology, which relies on in silico point mutations, reveals that DNNs trained for translation initiation site detection correctly identify well-established biological signals relevant to translation, including (1) the importance of the Kozak sequence, (2) the damaging consequences of ATG mutations in the 5’ untranslated region (UTR), (3) the detrimental effect of premature stop codons in the coding region, and (4) the relative insignificance of cytosine mutations for translation. Furthermore, we delve deeper into the Beta-globin gene and investigate various mutations that lead to the Beta thalassemia disorder. Finally, we conclude our work by laying out a number of novel observations regarding mutations and translation initiation. Availability For data, models, and code, visit github.com/utkuozbulak/mutate-and-observe Supplementary information Supplementary data are available at Bioinformatics online.

Assessing the reliability of point mutation as data augmentation for deep learning with genomic data

“…Consequently, splicing mutations can directly cause disease or influence disease susceptibility and severity. For instance, a single point mutation within the first intron of the beta-globin gene can cause beta thalassemia [ 6 ]. Thus, the interplay between splicing efficiency and intron removal is critical for maintaining proper gene expression and functionality.…”

“…In recent years, deep neural networks (DNNs) have emerged as highly effective models for addressing sequence-related problems. Notably, several prominent models have been developed for DNN-based detection of translation initiation sites (TIS), including TITER [ 4 ], NeuroTIS [ 49 ], DeepTIS [ 50 ], TISRover [ 5 ], and TISRover+ [ 6 ]. Additionally, for splice site detection, Deep belief networks [ 51 ], Spliceator [ 52 ], SpliceRover [ 7 ], and SpliceAI [ 8 ] were created.…”

“…For example, in the Bidirectional Encoder Representations from Transformers (BERT) model, masking is used to randomly conceal some of the input tokens and then train the model to predict the original tokens based on the context [ 37 ]. This technique of masking can be utilized in genomic data augmentation as well [ 6 , 38 ].…”

“…The early successes of DNNs in computer vision, for example in object recognition and image segmentation, demonstrated their ability to learn complex features from raw data and make accurate predictions [ 2 ]. These models have also shown great promise in the field of genomic data analysis, showcasing their ability to learn from and interpret large amounts of genetic data [ 3 ], and have seen increased use to solve a variety of biological problems including predicting translation initiation sites (TIS) [ 4 – 6 ], splice sites [ 7 , 8 ], promoter sites [ 9 ], functional effects of non-coding variants [ 10 ], and to characterize protein-specific properties [ 11 – 13 ]…”

“…In this work, we investigate the usage of single nucleotide polymorphisms (i.e., point mutations) as a data augmentation method for genomic data. Our investigation focuses on translation initiation site (TIS) detection [ 5 , 6 ], as well as splice site detection [ 7 ], since both are established as important tasks in genomics, cover both mutations in coding (i.e., exons) and non-coding (i.e., untranslated regions and introns) regions of genomic sequences, and are extremely sensitive to mutations [ 6 ]. Based on this investigation, we propose a principled and novel augmentation method that is straightforward to incorporate into any pipeline that employs such data.…”