Exploring De Novo Assembly of Protein Sequences from Simulated Reads Generated by a Single Molecule Protein Sequencing Technology

[awassie22]

Title

Exploring De Novo Assembly of Protein Sequences from Simulated Reads Generated by a Single Molecule Protein Sequencing Technology

Abstract

There are currently different approaches being explored for de novo single molecule protein sequencing (SMPS). While the proposed technologies differ in their sequencing mechanisms, at their core these approaches aim to sequence peptides or full-length proteins with single amino acid (AA) sensitivity and resolution. Such a high-throughput method for SMPS enables a wide-range of proteomic applications, one of which is the assembly of full length protein sequences from short, error-prone reads of peptides generated by the SMPS method. In the case of genomics, the early days saw the development of short-read assemblers for Solexa sequencers that enabled scalable whole genome sequencing. We foresee similar opportunities for protein sequencing that will be of great benefit to the proteomics community as SMPS technologies continue to mature over the next few years.
As part of this hackathon, we propose to explore and implement a strategy for a de novo assembler of full length protein sequences. To implement this strategy, we will provide a simulated dataset from a hypothetical SMPS technology. This SMPS sequencer has the ability to generate short reads from peptides. For each AA position, the sequencer will output posterior probabilities across the natural 20 AA, along with information on error rates

Project Plan

1. Propose a strategy for a de novo assembler of full length protein sequences. Or, as starting points, adapt existing protein sequence assemblers such as ALPS and PASS, which are based on weighted de Bruijn graph assembler and overlap extension approaches respectively.
2. Implement and test the strategy on simulated datasets from a hypothetical SMPS technology. To test performance in different contexts, these datasets will contain single molecule 10 aa-long reads of short peptides generated from purified mixtures of either 1 protein, 2 proteins, 10 proteins, or 100 proteins at a roughly equal abundance. (If longer than 10-aa reads are needed, longer simulated reads can be provided). These short peptides will be generated from a simulated digestion of the protein mixtures with multiple proteases in separate pools before being combined. The 10 aa long reads of the peptide fragments will have posterior probabilities at each AA position across the natural 20 AA. In addition, the sequencer readout will have a known deletion rate at each AA position. Each dataset will consist of a few million reads for the protein mixture scenarios; more reads can be provided as necessary.
3. When implementing the strategy, identify and develop metrics for the performance of the proposed assembler. For instance, accuracy of assembly, length of assembly, etc. In addition, identify key parameters of the assembler and their impact on these metrics.

After the hackathon we aim to continue further developing the proposed strategy, and ideally test the assembler with any SMPS dataset that becomes publicly available. We will make the code available via a public Github repository, and if a publication is produced we aim to make all contributors co-authors

Technical Details

• Programming Language: Python or other suitable language
• Existing Protein Sequence Assemblers: PASS and ALPS can serve as starting points, but we are not restricted by these existing methods and are open to explore new directions.
• Dataset: Simulated SMPS sequencing datasets of purified mixtures of human proteins will be generated and shared by us before the hackathon.

Contact information

Asmamaw (Oz) Wassie
Glyphic Biotechnologies
[email protected]

[magnuspalmblad]

This looks like a very interesting project! To make the enzymatic digestion simulation more realistic, we could consider using one of the deep learning models trained on mass spectrometry peptide identifications (though of course, they may also be biased toward what mass spectrometers can detect). There will be time (1.5 hours I think) for the project leads to present their topic, before people decide which project(s) to work on.