Awesome

Google Colaboratory notebooks for protein structure predictions for Discoba species

These Colaboratory notebooks are ideal for predicting structures of Trypanosoma and Leishmania species, along with other kinetoplastids, euglenids, diplonemids, etc.

To use a notebook, follow the link to the notebook code then click on the "Open in Colab" badge to run the notebook in Google Colab.

Get started

Go here to to generate an a3m alignment file using a HMMER search of the Discoba protein database, then use the official CoabFold notebook and upload the a3m file as a custom alignment to do your structure prediction.

Note that this generates an alignment based only on discoba sequences, not full eukaryotic diversity.

Multiple sequence alignment

These generate a multiple sequence alignment, generating an a3m alignment file suitable for use for use in other protein structure prediction pipelines.

DiscobaHMMER: For generating an a3m alignment file using a HMMER search of the Discoba protein database.

DiscobaMMSeqs2: For generating an a3m using an MMSeqs2 search of the Discoba protein database.

Discoba database generation

The transcriptomes folder contains the code necessary to fetch and build the predicted protein sequences of all Discoba species in the Discoba protein sequence database. Clone the repository, then run fetchAll.sh to build the entire database. This will take some time... Requires curl, gzip, perl, python, nodejs, cutadapt, jellyfish, bowtie2, bwa and samtools. Fetches and builds trinity, which requires autoconf, libbz2-dev and liblzma-dev. Fetches and builds cdhit which should require no further dependencies. Other tools are either scripts or fetched as precompiled packages.

Discoba database

The complete Discoba protein sequence database is available for download from Zenodo.org: discoba.fasta.gz and version information in discobaStats.txt. A mirrored copy is available via WheelerLab.net, which may be faster to download depending on where you are in the world: discoba.fasta.gz.

Protein structure predictions

Direct protein structure notebooks have been depreciated. The combination of Google Colab default package changes and ColabFold API changes make this too much of a moving target to effectively maintain.

DiscobaAlphaFold2HMMERv2: Experimental updated version for prediction of protein structures using the latest version of ColabFold modified to also include a HMMER search of the Discoba protein sequence database (recommended!).

DiscobaAlphaFold2HMMER: For prediction of protein structures using ColabFold modified to also include a HMMER search of the Discoba protein sequence database

DiscobaAlphaFold2MMSeqs2: For prediction using ColabFold modified to also include an MMSeqs2 search of the Discoba protein sequence database.

These protein structure predictions are based on ColabFold, which uses a MMSeqs2 search of UniRef and environmental sequence databases to make an MSA, then an AlphaFold2 structural prediction.

Acknowledgments

This work builds on the excellent ColabFold and would not be possible without the work by Sergey Ovchinnikov (@sokrypton), Milot Mirdita (@milot_mirdita) and Martin Steinegger (@thesteinegger).

Referencing

If you use structure predictions made using this resource, please cite:

Wheeler RJ. "A resource for improved predictions of Trypanosoma and Leishmania protein three-dimensional structure" PLoS One, doi: 10.1371/journal.pone.0259871 (2021)

Mirdita M, Ovchinnikov S and Steinegger M. "ColabFold - Making protein folding accessible to all." bioRxiv, doi: 10.1101/2021.08.15.456425 (2021)

Please also cite the relevant AlphaFold and database references, see ColabFold documentation for full information.

If you only use the multiple sequence alignments then please just cite Wheeler 2021.