Genopoeia

Mythological genomes for testing bioinformatics algorithms.

Design Goals

The initial design goal is to create a setting to evalute algorithms used for peak-finding. In the future, it may be used for alignment, spliced alignment, differential gene expression, variant detection, gene-prediction, or even pan-genomic analyses.

Genomes

A genome is a FASTA file of chromosomes and an associated GFF file that describes the features on the chromosomes. Genomes are generated pseudo-randomly using a seed, and the seed becomes part of the genome's name. g25 is a genome generated with a random seed of 25. All of its chromosome names start with g25. Chromosomes are numbered from 0 upwards. Smaller numbers are generally shorter and simpler chromosomes. For example, g25.0 is a short, simple chromosome made with seed 25.

Chromosomes, Loci, and Zones

Chromosomes are organized into descrete units with obvious bounaries. A chromosmal locus is defined as 10 kb. Each locus begins with 80 nts of Ns followed by 320 nt of unbiased sequence. Sequences with "interesting" content begin 401 bases into the locus and continue to 9600 bases into the locus. The last 400 bases of a locus are 320 nt of unbiased sequence followed by 80 nts of Ns.

An 'experimental zone' or just 'zone' is a collection of 100 loci, which corresponds to 1M. The smallest chromosomes are just 1 zone, but larger chromosomes may have several zones.

Genome Generation

Genomes are generated by a generalized hidden Markov model (GHMM) stored as a JSON file. The "name" of the GHMM should match the file name without the file extension. The "source" is a free text field that describes how the GHMM was built. Most of the GHMM is described as a dictionary of "state" objects.

Each "state" object is indexed by its name. State objects can be of three types: "markov", "pwm", or "static". A "markov" state generates sequence 1 nt at a time over a weighted random duration. This is used to generate features with variable lengths, such as intergenic sequences. A "pwm" state generates sequence from a position weight matrix. This is used for features with standard lengths and specific misspellings, such as splice sites. A "static" state generates an exact sequence and a corresponding "drift" rate. This is useful for representing repeats of various ages. All states also specify "init" and "term", which specifies the weight assoiciated with starting or ending in a state.

To see how all this fits together, it's best to look at a simple example.

{
	"name": "toy",
	"source": "fabricated",
	"states": {
		"iid": {
			"type": "markov",
			"init": 1,
			"term": 1,
			"transitions": {"hom": 1.0},
			"durations": [0, 0, 0, 1, 1, 1],
			"emissions": {"": [1, 1, 1, 1]},
		},
		"hom": {
			"type": "markov",
			"init": 0,
			"term": 0,
			"transitions": {"iid": 0.5, "rep": 0.5},
			"durations": [2, 2, 2, 1, 1, 1],
			"emissions": {
				"A": [9, 1, 1, 1],
				"C": [1, 9, 1, 1],
				"G": [1, 1, 9, 1],
				"T": [1, 1, 1, 9]
			}
		}
	}
}

The toy GHMM above flip-flops back and forth between two states: iid and hom. In the iid state, the emissions are unbiased and the states are 4-6 nt long. In the hom state, the emissions tend to be homopolymers, and the states are more likely to be 1-3 nt than 4-6.

Here are some more example states...

GHMM Files

Pre-built GHMMs are stored in the models directory.

Name	Description
iid	all nucleotides equally probable
ec.0	inspired by E. coli
ce.0	inspired by C. elegans

Repeat

A repeat is a sequence that shows up multiple times in a genome.

Bias

A special GFF feature, "bias" is used to describe various kinds of ascertainment bias. Bias can come from the sequence itself, or differences in gene expression. The feature "source" is "bias" and the feature "type" describes that bias in more detail.

• dna - the bias is seen when sampling DNA (e.g. ChIP seq)
• rna - the bias is seen when sampling RNA (e.g. differential gene expression)

The feature "score" describes the bias numerically, which is expressed in log2 units. A score of -1 has a weight of 0.5 (2**-1) while a score of 3 has a weight of 8 (2***3). The score of a bias feature is required (no dot). Here's an example bias feature from seed 25 on chromosome 000 with bias 2.1.

g25.000  bias  dna  2201  2300  2.1  .  .

Sampling

To sample sequencing reads from a genome, use sample_genome.py. This program uses various features of the genome.

• genome
• wide
• narrow
• mrna

Reconstructions and Subsets

Every genome is procedurally generated from a random seed. Some chromosomes of a genome might not be useful for a specific study. Therefore, it is often useful to create a subset of a genome, which is done with subset_genome.py.

Peak Feature

Gene Feature

Genes require more complex sequence model