Method

Method#

kaptive assembly#

For each input assembly, Kaptive runs the kaptive.assembly.typing_pipeline which does the following:

  1. Aligns locus gene nucleotide sequences to the assembly contig sequences using minimap2.

  2. Identifies the best matching locus type using the scoring algorithm.

  3. Extracts the locus gene sequences from the assembly contig sequences.

  4. Predicts the serotype/phenotype based on the gene content.

Scoring algorithm#

  1. A matrix is initialised with one row per locus and one column per score metric.

  2. For each gene found, the best alignment is chosen and the scores are added to the respective locus row if the coverage passes the threshold (--min-cov).

  3. The matrix is then weighted by the --weight-metric and the scores are selected from the column corresponding to the --score-metric.

  4. The top N loci (--n-best) are selected to be fully aligned to the assembly.

  5. Steps 1 and 2 are repeated and the best locus is selected from the column corresponding to the --score-metric.

The score metric can be explicitly set by the flag --score-metric, the options are:

  • 0 (AS) - Sum of the alignment scores per query

  • 1 (mlen) - Sum of the number of matching bases in the alignment per query

  • 2 (blen) - Sum of the number of aligned bases per query

  • 3 (q_len) - Sum of the number of bases of each query found (regardless of whether they are aligned)

The weighting can be explicitly set by the flag --weight-metric; the options are:

  • 0 - No weighting

  • 1 - Number of genes found

  • 2 - Number of genes expected

  • 3 - Proportion of genes found

  • 4 - Sum of the number of aligned bases per query (blen)

  • 5 - Sum of the number of bases of each query found (regardless of whether they are aligned) (q_len)

Note

The gene score matrix can be written to a TSV file using the --scores flag, however this will not type the assembly or reconstruct the locus.

Locus reconstruction#

After the best matching locus type has been identified, Kaptive will:

  1. For each contig, the ranges from the full-length locus alignments of the best match are extracted.

  2. The ranges are merged together if they are within the distance of the largest locus in the database.

  3. The merged ranges are used to create LocusPiece objects and the sequence is extracted from the assembly contig.

  4. Gene alignments are culled twice to determine the gene content:

    1. The first removes alignments overlapping genes from the best match.

    2. The remaining alignments that are not part of the best match are culled so that the best alignment is kept and each alignment represents the best gene for that part of the contig.

  5. For each remaining gene alignment, the gene is then evaluated.

Note

You may see multiple entries for the same gene in the gene details columns. This can happen when there are multiple significant alignments for a gene that has been split, either over contigs or by an artefact such as and insertion element. This will be evident by the low percent coverage of the entries, which should all add up to ~100% if they are indeed part of the same gene.

Gene evaluation#

For each GeneResult object, Kaptive will:

  1. Check whether the gene is partial by determining if the gene overlaps the contig boundaries.

  2. Extract the DNA sequence from the assembly contig and translate to amino acid.

  3. Perform pairwise alignment to the reference gene amino acid sequence and calculate percent identity.

  4. Check for truncation by determining if the amino acid sequence length is <95% of the reference gene protein length.

  5. Determine whether the gene belongs to the biosynthetic gene cluster (inside locus) or not (outside locus).

Note

Partial genes are not considered for truncation. This prevents false positive truncation calls in fragmented assemblies which may otherwise have an impact on phenotype prediction.

Phenotype prediction#

As of Kaptive 3, we have added the ability to predict the resulting phenotype of the assembly. This is similar to how the Type was reported in previous versions, but now includes the ability to predict specific phenotypes based on known mutations/modifications in a given set of locus genes as defined in the database logic file.

Confidence score#

Kaptive with finally calculate how confident it is in the prediction, which is explained here.