• Description of the work performed since the beginning of the project
Throughout 2018 I worked on two main projects: 1) development of a protocol for single cell isolation and sequencing from silkmoth pupae antennae in order to examine genetic correlates of OSN developmental diversification and wiring, and 2) development of a computational tool for chemosensory gene prediction in ant genomes in order to efficiently annotate these genes in hundreds of genomes for comparative evolutionary analyses. In early 2019, another research team successfully isolated and sequenced single cells from Drosophila antennae, publishing data and findings that rendered ongoing attempts to sequence single cells from silkmoth antennae redundant. This project was therefore abandoned. Consequently, I widened the scope of my work on ant genome annotation, and developed protocols for sequencing full-length transcripts from ants using cDNA amplification with cap-dependent second strand priming as well as Nanopore long read sequencing.
• Description of the main results achieved so far
In order to accurately identify chemosensory genes in ant genomes, I developed the HAP.py-ABCENTH pipeline and prediction program (
https://github.com/biorover/HAPpy-ABCENTH(si apre in una nuova finestra)). This program consists of two components: HAP.py (Homology Annotation Pipeline) and ABCENTH (Annotation Based on Conserved Exons Noticed Through Homology). HAP.py is a wrapper for a variety of homology based gene finding tools. In the context of annotating ant chemosensory genes, HAP.py is used to cluster a set of reference genes into closely related groups, each of which has completely conserved exon number and intron phases in most Hymenoptera chemosensory receptor gene families (ORs and GRs). HAP.py then uses the HMMER suite to build hidden markov model (HMM) profiles for each exon for each cluster and search them against translated ORFs from a target genome. From here, exon structure and HMMER homology search hits for each gene cluster are fed into ABCENTH, which tries to extend (or sometimes trim) each exon candidate to valid splice motifs that are the most consistent with the exon’s putative size and splicing phase. ABCENTH then connects all valid exons arranged in the proper order. ABCENTH will also find exons missed during the homology search by extracting ORFs between splice motifs with the proper phase and roughly the proper size and comparing these to the HMM of the missing exon to identify the best candidate. This feature allows for the identification of short exons often missed by purely homology-based pipelines. Application of the HAP.py-ABCENTH pipeline to previously published ant genomes showed that on average it achieved over 98% recall and over 95% precision at the complete gene level (meaning only perfectly predicted genes counted as true positives; see attached figure) when compared with manual annotations.
In order to generate a set of perfectly annotated genes to benchmark HAP.py-ABCENTH as well as to serve as a resource for development of other gene prediction tools for ant genomes, I sequenced several million full length transcripts from five species of ants using poly-A primed mRNA reverse transcription with cap-dependent second-strand synthesis followed by Oxford Nanopore long read sequencing. Reads were QCed, filtered, clustered, and polished using the PyChopper, Pinfish, and TALC software, yielding approximately 300,000 unique cleaned transcripts for each species. Transdecoder analysis and clustering into genes based on overlapping coding sequences showed that the transcripts came from between 8,600 and 9,600 genes per species. Full length transcripts of chemosensory genes matched the gene structure predicted from reference annotations in over two dozen manually inspected cases. Preliminary evaluation of genome wide gene prediction based on training with conserved genes identified through this pipeline suggested that these models may be very useful for high throughput gene prediction training.
