aTRAM is an iterative assembler that performs reference-guided local de novo assemblies using a variety of available methods. It is well-suited to various tasks where Next-Generation Sequencing (NGS) data needs to be queried for gene sequences, such as phylogenomics. It is actually a suite of programs:
-
atram_preprocessor.py: Builds a set of databases from your input NGS sequences. One is an SQLite3 database that holds the original NGS sequences. It also creates a set of BLAST databases for finding sequence matches.
-
atram.py: Uses the databases from step 1 along with a reference sequence (aka a bait sequence) and a de novo assembler program. It's the part of aTRAM that actually builds the assemblies.
-
atram_stitcher.py: Takes the assemblies from atram.py and joins them together. It also uses an iterative approach and the Exonerate program.
This program reads a series of related FASTA or FASTQ files and builds an aTRAM database. This aTRAM database is actually two or more databases (typically several).
-
An SQLite3 database that holds the contents of the FASTA/FASTQ files in a format that can be easily and quickly queried. It takes three pieces of information from the original files are: the sequence name, sequence end (1, 2, or none), and the sequence itself.
-
A set of BLAST databases. atram.py uses multiple BLAST databases. This dataset division enables parallelized read queries and greatly improves performance even for serial queries.
-
Note: That during the assembly process, in atram.py, other temporary SQLite3 and BLAST databases will also be built.
This is the heart of aTRAM. This program performs the actual gene assembly. There are many options and knobs for atram.py but the basic algorithm is:
BLAST the input bait sequence against the BLAST databases created by atram_preprocessor.py. This will yield a subset of your original FASTA/Q input sequences that match the bait sequence.
For the first iteration we choose what to blast against. Either DNA or amino acid sequences. Using amino acid sequences allows you to look for more distantly related taxa. All subsequent iterations will use the output from the assembler. The input bait sequences are in FASTA/Q format.
Now that we have a set of BLAST hits and we want to assemble contigs out of them but we also want to consider any matching ends during the new assembly. That is, if an end 1 matches we pull in the end 2 and vice versa. We know that both ends come from the same sequence but only one end might be covered the target sequence we want to pull in the other end to help with the assembly process.
Contigs are built up using any one of the following assemblers: Velvet, Trinity, Abyss, or Spades. Further, because the reads are assembled de novo, the assemblies are not as tightly restricted to the reference sequence. Therefore, inversions or other structural differences in the newly sequenced genome will be revealed.
Assemblies are improved by an iterative approach: in the second iteration, the assembled contigs replace the original query, and are blasted against the short read database. Matching reads are then assembled de novo as in the first iteration. This process continues until the user-specified iteration limit is reached or no new contigs are assembled.
This program takes the contigs assembled by atram.py and stitches them into longer sequences.
Given:
- The assembled contigs from atram.py
- a text file of all your taxon names
- a fasta file of all the reference genes in amino acids
- the program exonerate - https://www.ebi.ac.uk/about/vertebrate-genomics/software/exonerate-user-guide
The taxon and gene names must correspond to the assembly file names. For example the assemblies from atram2.0 will produce a fasta file with the gene name followed by the taxon name. The taxon names and the gene names in the amino acid file must correspond.
This program will group all of the taxa for each gene together. It will use the program exonerate and your amino acid reference file to find exon positions in the assemblies. It will then stitch those exons together. If there is a missing exon then the script will add groups of 3 NNNs in the missing places so that in the end, there will be 1 file for each gene with all of the exons for each taxon and with NNNS in the missing pieces. Therefore all of the exon lengths for each taxa are roughly the same, barring indels and ready for alignment steps.
