Skip to content

Latest commit

 

History

History
42 lines (27 loc) · 2.94 KB

README.md

File metadata and controls

42 lines (27 loc) · 2.94 KB
Raw

Analyzing RNA secondary structure prediction algorithms

This repository evaluates different secondary structure prediction algorithms. The basic workflow is:

  • The input is a sequence of RNA nucelotides (and in some cases, other inputs as well)
  • We pass that input into a model (i.e., RNAfold, EternaFold, SPOT-RNA, etc.)
  • The model outputs a secondary structure prediction based on the input
    • This output could come in several forms, but we want the dot-bracket notation
    • If the output is in anything besides a dot-bracket string, convert it to db somehow (like in the case of SPOT-RNA)
  • Evaluate the prediction's accuracy
    • So far, we do this by comparing experimentally-derived reactivities of RNA structures with the output of the dot-bracket notation
    • There are other ways to evaluate accuracy but with the ss_deeplearning_data repository, this is the best evaluation so far

Once more prediction algoirthms are identified and support for their functionality are added, I intend to build a database of evaluations. The databse will contain

  • Evaluation of each algorithm and for each datapoint
  • A ranking of the "best" and "worst" algorithms
  • A list of common structures that are incorrectly guessed

Code organization

This explains the different code files and will grow as time goes

image

models

DataPoint

Scorer

In this module, we put different scoring algorithms. For example, the DSCI scorer--which uses the Mann-Whitney U-test--is encapsulated in the DSCI class. All classes in this module must inherit the base Scorer class and override the evaluate method. Note that the Scorer class is instantiated with a DataPoint object, algorithm name (e.g., EternaFold), and an optional parameter called evaluate_immediately. The evaluate_immediately parameter defaults to True and is used to indicate if the user wants the secondary structure prediction to be scored as soon as the Scorer object is instantiated. In other words, when evaluate_immediately is True, the evaluate method will be called at the end of the __init__ method. If not, the user will have to manually call the evaluate method on the Scorer object.

predictors

This file contains classes that encapsulate different ways in which the algorithms must be executed and evaluated. For example, the Eterna class calls the prediction method for EternaFold on a given sequence file. Conversely, the SPOT_RNA class executes the prediction algoirthm, but also converts the output .ct files into a secondary structure dot-brakcet string so the predictions can be evaluated.

script.py

This file simply exists to test and show examples of how to use the executions models and evaluations.

TODO

This project has a long way to go. Currently, some of the work remains to standardize inputs and make it painfully obvious how users can add new tools.