AutoTuneX

This repo contains the trained parameter advisor models for Scallop and StringTie2. For details on these models, please read our manuscript "Data-driven AI system for learning how to run transcript assemblers" (https://doi.org/10.1101/2024.01.25.577290).

Installation

Install the python dependencies: pip install -r requirements.txt, we tested on python>=3.9.

Usage

Here we provide an example of generating a parameter advisor set for a new RNA-seq sample via our model.

Download fastq files

Download fastq files of the sample (e.g. SRA accession number:SRR1023790) from https://sra-explorer.info.

> srapath SRR1023790
https://sra-pub-run-odp.s3.amazonaws.com/sra/SRR1023790/SRR1023790
> wget https://sra-pub-run-odp.s3.amazonaws.com/sra/SRR1023790/SRR1023790
> fastq-dump-orig --split-files --gzip --outdir ./ SRR1023790

(fastq-dump-orig is not available in sratoolkit.3.1.1-ubuntu64, use fastq-dump instead.)

Generate MinHash sketch

1. Download Mash from https://mash.readthedocs.io/en/latest/.
2. Run the command:

cat SRR1023790_1.fastq.gz SRR1023790_2.fastq.gz | ./mash-Linux64-v2.3/mash sketch -r -m 2 -o SRR1023790 -

It will output SRR1023790.msh: MinHash sketch of the sample SRR1023790.

Generate advisor set

1. download all the files from https://doi.org/10.1184/R1/25037156.v1, here we need to use the following files:

   (1) trained_scallop.pth: the trained parameter advising model for Scallop.

   (2) trained_stringtie.pth: the trained parameter advising model for StringTie2.

   (3) train_features_withaug_scallop.npy: the set representations of all the samples in the representative set of Scallop, including all the samples from the data augmentation module.

   (4) train_features_withaug_stringtie.npy: the set representations of all the samples in the representative set of StringTie2, including all the samples from the data augmentation module.

2. Put the files described above, all the files from the folder ./files/, and the files autoparadvisor_contrastive.py, MinHash.capnp, advisorset_generator.py as well as MinHash sketch SRR1023790.msh into the same folder, run the command:

python advisorset_generator.py --name SRR1023790 --assembler scallop --top 5

or

python advisorset_generator.py --name SRR1023790 --assembler stringtie --top 5

Here the value of --top is the size of the advisor set. top>=5 is recommended.

3. Our script will output a folder ./SRR1023790_scallop_advisorset/ (or ./SRR1023790_stringtie_advisorset/) and all the parameter candidates are stored there.

Other files

(1) representative_sample_scallop: accession numbers of 1263 representative samples for Scallop.

(2) representative_sample_stringtie: accession numbers of 1235 representative samples for StringTie2.