Assessing the Use of Secondary Structure Fingerprints and Deep Learning to Classify RNA Sequences
Bibliographic record
Abstract
Non-coding RNAs (ncRNAs) are RNA molecules that do not code for protein, but take part in biological processes, including gene expression. Interestingly, like proteins, they can fold into complex structures to perform their wide array of biological functions. Since the folded structure of a ncRNA may be critical to its function, many studies have attempted to exploit structural data to infer information, often using machine learning techniques. For instance, they have used predicted secondary structures as input features to various machine learning techniques, in order to classify RNA sequences. However, it is known that a strand of RNA can fold into more than one possible structure, and some strands even form different structures in vivo and in vitro. Furthermore, ncRNAs often function as RNA-protein complexes, which can affect structure. We therefore hypothesized that using a single predicted secondary structure for a single sequence may discard important information, which may result in poorer classification accuracy. To investigate this claim, we propose the use of secondary structure fingerprints as features for machine learning applications, and report on a preliminary evaluation of this approach. The fingerprints comprise two categories: a higher-level (topological) representation derived from RNA-As-Graphs (RAG), and free energy fingerprints based on a novel curated repertoire of small RNA motifs. We have also evaluated our deep learning architecture with k-mers as features, alone and combined with secondary structure fingerprints; to see whether secondary structures or nucleotide composition is more useful in RNA classification, and whether or not both feature types complement each other well. The dataset, trained models, and supplemental material of this study are available at https://www.site.uottawa.ca/turcotte/bibm2020.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.010 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.002 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".