#bioinformatics #rna #structural-alignment

bin+lib consprob

Quick Probability Inference Engine on RNA Structural Alignment

18 releases

Uses old Rust 2015

0.1.17 Nov 23, 2022
0.1.15 Aug 4, 2022
0.1.14 Jul 26, 2022
0.1.11 Mar 16, 2022
0.1.0 Jul 13, 2020

#80 in Science

Download history 12/week @ 2022-08-15 29/week @ 2022-08-22 10/week @ 2022-08-29 56/week @ 2022-09-05 52/week @ 2022-09-12 96/week @ 2022-09-19 87/week @ 2022-09-26 123/week @ 2022-10-03 44/week @ 2022-10-10 22/week @ 2022-10-17 48/week @ 2022-10-24 59/week @ 2022-10-31 41/week @ 2022-11-07 46/week @ 2022-11-14 72/week @ 2022-11-21 33/week @ 2022-11-28

196 downloads per month
Used in 4 crates (3 directly)

MIT license

1.5MB
3K SLoC

Quick Probability Inference Engine on RNA Structural Alignment

Installation

This project is written in Rust, a systems programming language. You need to install Rust components, i.e., rustc (the Rust compiler), cargo (the Rust package manager), and the Rust standard library. Visit the Rust website to see more about Rust. You can install Rust components with the following one line:

$ curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

Rustup arranges the above installation and enables to switch a compiler in use easily. You can install ConsProb:

$ # AVX, SSE, and MMX enabled for rustc (another example: RUSTFLAGS='--emit asm -C target-feature=+avx2 -C target-feature=+ssse3 -C target-feature=+mmx -C target-feature=+fma')
$ RUSTFLAGS='--emit asm -C target-feature=+avx -C target-feature=+ssse3 -C target-feature=+mmx' cargo install consprob

Check if you have installed ConsProb properly:

$ consprob # Its available command options will be displayed

You can run ConsProb with a prepared test set of sampled tRNAs:

$ git clone https://github.com/heartsh/consprob && cd consprob
$ cargo test --release -- --nocapture

Structural Alignment Scoring Models

While ConsProb's paper describes only the Turner 2004 model as an available scoring model of RNA structural alignment, ConsProb also offers the CONTRAfold v2.02 model. These two scoring models are described here. The CONTRAfold v2.02 model comprises fewer feature scoring parameters than the Turner 2004 model. My prediction accuracy benchmarking of ConsAlifold adopting ConsProb showed the marginal difference between the CONTRAfold v2.02 model and the Turner 2004 model:

Structure prediction accuracy comparison
F1 score-based p-value MCC-based p-value
2.9 * 10-5 1.2 * 10-5

In my running time benchmarking of ConsAlifold adopting ConsProb, the CONTRAfold v2.02 model was slightly slower than the Turner 2004 model due to the larger spaces of possible RNA structural alignments:

Prediction running time comparison

Advanced Computation of RNA Structural Context Profiles

Measuring the structural context profile of each RNA nucleotide (i.e., the posterior probability that each nucleotide is in each structural context type) is beneficial to various structural analyses around functional non-coding RNAs. For example, CapR computes RNA structural context profiles on RNA secondary structures, distinguishing (1) unpairing in hairpin loops, (2) base-pairings, (3) unpairing in 2-loops (e.g., bulge loops and interior loops), (4) unpairing in multi-loops, and (5) unpairing in external loops as available structural context types:

CapR's structural context profiles

Respecting CapR, ConsProb offers the computation of average structural context profiles on RNA structural alignment, distinguishing the above structural context types. Technically, ConsProb calculates the structural context profile of each nucleotide pair on RNA pairwise structural alignment and averages this pairwise context profile over available RNA homologs to each RNA homolog, marginalizing these available RNA homologs. ConsProb's context profile computation is available for the Turner 2004 model and the CONTRAfold v2.02 model but is not described in ConsProb's paper. (You can easily derive this context profile computation by customizing ConsProb's main inside-outside algorithm for computing posterior nucleotide pair-matching probabilities, as CapR is based on McCaskill's algorithm.) The below is examples of ConsProb's average context profiles:

ConsProb's average context profiles

Docker Playground

Replaying computational experiments in academic papers is the first but troublesome step to understand developed computational methods. I provide an Ubuntu-based computational environment implemented on Docker as a playground to try out ConsProb:

$ git clone https://github.com/heartsh/consprob && cd consprob
$ docker build -t heartsh/consprob .

You can dive into the Docker image "heartsh/consprob" built by the above commands, using Zsh:

$ docker run -it heartsh/consprob zsh

Method Digest

LocARNA-P can compute posterior nucleotide pair-matching probabilities on RNA pairwise structural alignment. However, LocARNA-P simplifies scoring possible pairwise structural alignments by utilizing posterior nucleotide base-pairing probabilities on RNA secondary structures. In other words, LocARNA-P does not score possible pairwise structural alignments at the same level of scoring complexity as many RNA folding methods. More specifically, many RNA folding methods such as RNAfold score possible RNA secondary structures distinguishing RNA loop structures, whereas many structural alignment-based methods such as LocARNA-P score possible pairwise structural alignments ignoring RNA loop structures. As an antithesis to these structural alignment-based methods, I developed ConsProb implemented in this repository. Distinguishing RNA loop structures, ConsProb rapidly estimates various pairwise posterior probabilities, including posterior nucleotide pair-matching probabilities. ConsProb summarizes these estimated pairwise probabilities as average probabilistic consistency, marginalizing multiple RNA homologs to each RNA homolog.

Author

Heartsh

License

Copyright (c) 2018 Heartsh
Licensed under the MIT license.

Dependencies

~28MB
~200K SLoC