Getting Started

git clone https://github.com/lh3/fermi-lite
cd fermi-lite && make
./fml-asm test/MT-simu.fq.gz > MT.fq
# to compile your program:
gcc -Wall -O2 prog.c -o prog -L/path/to/fermi-lite -lfml -lz -lm -lpthread

Introduction

Fermi-lite is a standalone C library as well as a command-line tool for assembling Illumina short reads in regions from 100bp to 10 million bp in size. It is largely a light-weight in-memory version of fermikit without generating any intermediate files. It inherits the performance, the relatively small memory footprint and the features of fermikit. In particular, fermi-lite is able to retain heterozygous events and thus can be used to assemble diploid regions for the purpose of variant calling. It is one of the limited choices for local re-assembly and arguably the easiest to interface.

If you use fermi-lite in your work, please cite the FermiKit paper:

Li H (2015) FermiKit: assembly-based variant calling for Illumina resequencing data, Bioinformatics, 31:3694-6.

Usage

For now, see example.c for the basic use of the library. Here is a sketch of the example:

#include <stdio.h>                      // for printf()
#include "fml.h"                        // only one header file required

int main(int argc, char *argv[])
{
	int i, n_seqs, n_utgs;
	fml_seq1_t *seqs;                   // array of input sequences
	fml_utg_t *utgs;                    // array of output unitigs
	fml_opt_t opt;
	if (argc == 1) return 1;            // do nothing if there is no input file
	seqs = fml_seq_read(argv[1], &n_seqs); // or fill the array with callers' functions
	fml_opt_init(&opt);                 // initialize parameters
	utgs = fml_assemble(&opt, n_seqs, seqs, &n_utgs); // assemble!
	for (i = 0; i < n_utgs; ++i)        // output in fasta
		printf(">%d\n%s\n", i+1, utgs[i].seq);
	fml_utg_destroy(n_utgs, utgs);      // deallocate unitigs
	return 0;
}

The fml_assemble() output is in fact a graph. You may have a look at the fml_utg_print_gfa() function in misc.c about how to derive a GFA representation from an array of fml_utg_t objects.

Overview of the Assembly Algorithm

Fermi-lite is an overlap-based assembler. Given a set of input reads, it counts k-mers, estimates the k-mer coverage, sets a threshold on k-mer occurrences to determine solid k-mers and then use them correct sequencing errors (Li, 2015). After error correction, fermi-lite trims a read at an l-mer unique to the read. It then constructs an FM-index for trimmed reads (Li, 2014) and builds a transitively reduced overlap graph from the FM-index (Simpson and Durbin, 2010; Li, 2012), requiring at least l-bp overlaps. In this graph, fermi-lite trims tips and pops bubbles caused by uncorrected errors. If a sequence in the graph has multiple overlaps, fermi-lite discards overlaps significantly shorter than the longest overlap -- this is a technique applied to overlap graph only. The graph after these procedure is the final output. Sequences in this graph are unitigs.

Limitations

1. Fermi-lite can efficiently assemble bacterial genomes. However, it has not been carefully tuned for this type of assembly. While on a few GAGE-B data sets fermi-lite appears to work well, it may not compete with recent mainstream assemblers in general.

2. Fermi-lite does not work with genomes more than tens of megabases as a whole. It would take too much memory to stage all data in memory. For large genomes, please use fermikit instead.

3. This is the first iteration of fermi-lite. It is still immarture. In particular, I hope fermi-lite can be smart enough to automatically figure out various parameters based on input, which is very challenging given the high variability of input data.