Background: The analysis of next-generation sequencing data from large genomes is a timely research topic. Sequencers are producing billions of short sequence fragments from newly sequenced organisms. Computational methods for reconstructing sequences (whole-genome assemblers) are typically employed to process such data. However, one of the main drawback of these methods is the high memory requirement. Results: We present Mapsembler, an iterative targeted assembler which processes large datasets of reads on commodity hardware. Mapsembler checks for the presence of given regions of interest in the reads and reconstructs their neighborhood, either as a plain sequence (consensus) or as a graph (full sequence structure). We introduce new algorithms to retrieve homologues of a sequence from reads and construct an extension graph. Conclusions: Mapsembler is the rst software that enables de novo discovery around a region of interest of gene homologues, SNPs, exon skipping as well as other structural events, directly from raw sequencing reads. Compared to traditional assembly software, memory requirement and execution time of Mapsembler are considerably lower, as data indexing is localized. Mapsembler can be used at http://mapsembler.genouest.org