Congrats Martin! 🥳🥳

Thanks Roland! It’s good to mention here that Hairsplitter is also part of it 😄

Thanks! 😊

Work with @msikic.bsky.social, @rvicedomini.bsky.social, Kresimir Krizanovic
MADRe is open-source, modular, and ready to use.
Check it out:
🔗 github.com/lbcb-sci/MADRe
9/9

A key feature of MADRe is its focus on organisms with sufficient abundance to be assembled.
While low-abundance strains may be underrepresented, this trade-off significantly reduces false-positive identifications, a common issue in strain-level metagenomics.
8/9

We evaluated MADRe on both real and simulated datasets and observed:
✅ Comparable or improved accuracy over existing tools
✅ Clearer and more realistic abundance profiles
✅ Substantial reductions in runtime and memory usage
7/9

While assembly is often considered computationally expensive, we demonstrate that MADRe, by combining assembly with contig-level mapping, is more efficient than directly mapping large volumes of reads to a full reference database. 6/9

To complete the pipeline, MADRe maps reads to the reduced reference database and applies a second round of probabilistic reassignment.
This enhances classification sensitivity and filters false-positive identifications, enabling precise strain-level profiling. 5/9

This reduction identifies candidate genomes present in the sample.
However, this step alone does not eliminate all false positives and does not provide accurate abundance estimates. 4/9

MADRe begins by assembling the metagenomic sample and mapping the resulting contigs - often representing collapsed strains - to a (large) reference database.
Using EM-based read reassignment and info about strain collapses, we construct reduced database. 3/9

Strain-level classification requires large reference databases, especially when there is no prior knowledge about sample composition.
However, mapping reads to such large databases is computationally expensive and often impractical at scale. 2/9