LMAS

Inês Mendes

MRamirez Lab - iMM

@ines_cim

cimendes

Metagenomic Assembly

Towards Accreditation in Metagenomics for Clinical Microbiology

Advisor Comittee Meeting

14 July 2021

Microbial (meta)genomics

Shotgun Sequencing Data

Direct Read Classification Methods

Assembly-Based Methods

Marker Gene Detection

Taxonomical Assignment

Abundance estimation

Sequence binning

Abundance estimation

Computational expensive
Lacks context

De novo metagenomic assemby

De novo metagenomic co-assemby

Consensus metagenomic assemby

Provides contextual information
Normalizes information

Microbial (meta)genomics

| Assembly

The assembly methods provide longer sequences that are more informative than shorter sequencing data and can provide a more complete picture of the microbial community in a given sample.

Assembly

consensus

de novo

OLC

de Bruijn graph

Microbial (meta)genomics

| Assembly

Results are highly dependent on the tools chosen for the analysis - Lack of standardization and proper benchmark.

Major issues

Highlights the potential and the limitations of shotgun metagenomics as a diagnostic tool - Lack of reproducibility

Reads

Contigs

Genomes

Microbial (meta)genomics

| Assembly

State-of-the-art software assessment for short-read metagenomic de novo assembly (Task 1)

Main goals

Evaluation of the influence the choice of assembler has on quality, reproducibility and resource requirements for metagenomic analyisis

Development of a reproducible, user-friendly workflow for the assessment of assembly success (Task 2)

Reads

Contigs

Genomes

Microbial (meta)genomics

| Assembly

Limited to short-read paired-end sequencing data

Main limitations

Reads

Contigs

Genomes

Limited to de novo assembly software

Lack of flexibility for the introduction of new tools

7 months ago...

Compiled a collection of de novo assembly tools, including Overlap, Layout and Consensus (OLC) and De Bruijn graph (dBg) assembly algorithms, with both single k-mer and multiple k-mer value approaches, and hybrid assemblers. The collection includes both genomic and metagenomic assemblers.

7 months ago...

Initial development of global and reference specific performance metrics for the assessment of de novo assembly quality.

7 months ago...

https://github.com/cimendes/metagenomic-assembler-comparison

Reference Dataset (Complete Bacterial Genomes)

In silico mock sample (even)

In silico mock sample (log)

Zymos standard (even)

Zymos standard (log)

3.7 M read pairs

8.8 M read pairs

47.8 M read pairs

Basic Assembly Workflow

Assembly Quality Assessment

https://github.com/cimendes/LMAS

https://lmas.readthedocs.io/

| Last Metagenomic Assembler Standing

Automated workflow enabling the benchmarking of genomic and metagenomic prokaryotic de novo assembly software using defined mock communities.

LMAS

| Last Metagenomic Assembler Standing

The input data is assembled in parallel by the set of genomic and metagenomic de novo assemblers in LMAS.

The global and per reference metrics are grouped in the interactive LMAS report for exploration.

The resulting assembled sequences are processed and assembly quality metrics are computed.

LMAS

| Last Metagenomic Assembler Standing

LMAS

| Last Metagenomic Assembler Standing

LMAS requires a Nextflow installation (version ≥ 21.04.1), requiring BASH and Java 8 (or higher).

All components of LMAS are executed in containers.

Continuous integration of the python templates for the quality assessment of assemblies is performed with pytest and GitHub Actions.

LMAS can be installed through Github.

LMAS

| Assembly Quality Metrics

The tabular presentation allows direct comparison of exact values between assemblies, and the interactive plots allow for an intuitive overview and easy exploration of results.

LMAS

| Report

LMAS

| Report

The results are presented in an interactive HTML composed of two main panels:

- a top summary panel with information on input samples and the LMAS execution,

- a bottom panel where selected global and reference specific performance metrics can be explored for each sample.

The JavaScript source code for the interactive report comes bundled with LMAS but is freely available at https://github.com/cimendes/lmas_report.

Eight bacterial genomes
Four plasmids
Even and logarithmic distribution of species
Simulated samples (with and without error) - InSilicoSeq

Sample	Distribution	Error Model	Read Pairs (M)
ENN	Even	None	8.6
EHS	Even	Illumina HiSeq	8.6
ERR2984773	Even	Real Sample	8.6
LNN	Log	None	47.5
LHS	Log	Illumina HiSeq	47.5
ERR2935805	Log	Real Sample	47.5

LMAS

| ZymoBIOMICS Microbial Community Standards

LMAS

| ZymoBIOMICS Microbial Community Standards

LMAS

| ZymoBIOMICS Microbial Community Standards

Some assemblers perform poorly
Resource usage is different
Assembler performance is influenced by species abundance and sample composition
Certain genomic regions are problematic for all assemblers
No one size fits all assembler!

LMAS

| 3 LMAS runs

https://github.com/cimendes/LMAS_Zymos_Resuls

LMAS

| 3 LMAS runs

There’s a disparity in usage for the evenly and logarithmically distributed samples, with the latter having more resource-intensive requirements. The resource usage also varied greatly by assembler, with multiple k-mer dBG having overall higher resource usage.

LMAS

| 3 LMAS runs

Only GATBMiniaPipeline, IDBA-UD and Minia assemblers produced inconsistent contigs (0.22%, 0.04% and 0.12% of the total contigs produced by the assemblers, respectively)

LMAS

| 3 LMAS runs

The logarithmically distributed samples showed greater variation than the evenly distributed counterparts, and the real samples showed greater variation than the mock samples.

The greatest difference observed was in the total basepairs produced by the assemblers but it didn’t impact the contiguity of the assemblies (N50) or the maximum contig size obtained.

After filtering for the minimum contig length of 1000 basepairs, the effect of the variation in assembly length is mitigated, with all assemblers showing increased robustness in the number of contigs and total basepairs assembled.