Forward’s database

Relational databases are automatically built to hold analysis results for any experiment. This is done using SQLAlchemy an Object Relational Mapper for Python. This package makes it easy to use one of the many supported RDBMS backends while writing technology agnostic code. It also maps database entries to Python objects which makes the integration seamless and facilitates some technical aspects.

This section describes the different Python classes that will get translated to database tables.

Variant table

class forward.genotype.Variant(**kwargs)

ORM Variant object.

Column	Description	Type
name	The variant’s name (e.g. rs12356)	String(25)
chrom	The chromosome (e.g. ‘3’ or ‘X’)	String(15)
pos	The variant’s position on the chromosome	Integer
mac	Minor allele count (e.g. 125)	Integer
minor	Least common allele (e.g. ‘T’)	String(10)
major	Most common allele (e.g. ‘C’)	String(10)
n_missing	Number of missing genotypes for this variant	Integer
n_non_missing	Number of non-missing genotypes for this variant	Integer

Computed fields:

• maf: mac / (2 \(\cdot\) n non missing)
• completion_rate: n non missing / (n non missing + n missing)

Results tables

class forward.experiment.ExperimentResult(**kwargs)

SQLAlchemy class to handle experimental results.

Column	Description	Type
pk	The primary key (autoincrement)	Integer
tested_entity	Either ‘variant’ or ‘snp-set’ depending on the test.	Enum
results_type	Polymorphic identity to identify (default: ‘GenericResults’)	String(25)
task_name	Name of the task	String(25)
entity_name	Variant or snp set identifier	String(25)
phenotype	Tested outcome	String(30)
significance	Significance value (e.g. p-value)	Float
coefficient	Effect coefficient (e.g. beta)	Float
test_statistic	Statistical test statistic (e.g. t-test statistic)	Float
standard_error	Standard error of the coefficient	Float
confidence_interval_min	Lower bound of the 95% CI on the coefficient	Float
confidence_interval_max	Higher bound of the 95% CI on the coefficient	Float

Todo

This could potentially be refactored so that tested_entity is a foreign key to the forward.genotype.Variant class. This would require using concrete table inheritance.

class forward.tasks.LinearTestResults(**kwargs)

Table for extra statistical reporting for linear regression.

Column	Description	Type
pk	The primary key, the same as the experiment.ExperimentResult	Integer
adjusted_r_squared	The adjusted R squared as reported by statsmodels	Float
std_beta	The standardized effect size. This is for \(x, y \sim \mathcal{N}(0,1)\)	Float
std_beta_min	Lower bound of the 95% CI for the standardized Beta	Float
std_beta_max	Higher bound of the 95% CI	Float

It is interesting to report the standardized beta to easily compare the effect size between different outcomes (that have different units). We will also report the coefficient of determination (R^2) that reports the fraction of explained variance.