. 2021 Feb 1;27(1):38–48. doi: 10.1038/s41380-021-01031-2

Table 2.

Summary of the 3 P-factors and the methods by which they were derived.

	Statistical methods with relevant information	Design	Psychopathology Assessments	Key references [1]
Phenotypic P-factor	Most common methods: SEM, factor analysis These are variance decomposition methods and path diagrams within the phenotypic domain. SEM and factor analysis are generally exploratory of multiple a priori models and/or a test of predefined factor structure	• Generally population based • High-risk cohort [13]	• Generally continuous symptom ratings • Binary diagnosis [12] • Probability of diagnosis [13]	[1, 2, 11–16]
Genomic P-factor	Most common methods: PCA, Genomic SEM These are variance decomposition methods on the genetic correlations across disorders, which are derived using LDSR of GWAS stats [3], SNP-based methods [4], or family/twin-based approaches. PCA is fully data-driven, while Genomic SEM is exploratory or confirmatory of a hypothesized factor structure.	Case-control GWAS designs	Clinical diagnosis	[3, 4, 27, 102]
Neural P-factor	Most common methods: CCA, ICA CCA can be used to extract “modes” across behavioural/environmental and MRI measures simultaneously. A mode is driven by a combination of the correlations within and across the variable classes. ICA tends to be run within the MRI domain. The resultant “components” can then be correlated with the behavioural and environmental variables. Both ICA and CCA are fully data-driven and are useful to extract a large number of “factors” from high-dimensional data.	• Population based, adults (Human Connectome Project) • Children and adolescent population, enriched for mental distress (ABCD Cohort)	• Behavioural, demographic & environmental measures • Self-report of (family) diagnosis and substance use • Cognitive task performance	[5, 6, 39, 43]

Statistical methods with relevant information

Design

Psychopathology Assessments

Key references [1]

Phenotypic P-factor

Most common methods: SEM, factor analysis

These are variance decomposition methods and path diagrams within the phenotypic domain. SEM and factor analysis are generally exploratory of multiple a priori models and/or a test of predefined factor structure

• Generally population based

• High-risk cohort [13]

• Generally continuous symptom ratings

• Binary diagnosis [12]

• Probability of diagnosis [13]

[1, 2, 11–16]

Genomic P-factor

Most common methods: PCA, Genomic SEM

These are variance decomposition methods on the genetic correlations across disorders, which are derived using LDSR of GWAS stats [3], SNP-based methods [4], or family/twin-based approaches.

PCA is fully data-driven, while Genomic SEM is exploratory or confirmatory of a hypothesized factor structure.

Case-control GWAS designs

Clinical diagnosis

[3, 4, 27, 102]

Neural P-factor

Most common methods: CCA, ICA

CCA can be used to extract “modes” across behavioural/environmental and MRI measures simultaneously. A mode is driven by a combination of the correlations within and across the variable classes.

ICA tends to be run within the MRI domain. The resultant “components” can then be correlated with the behavioural and environmental variables.

Both ICA and CCA are fully data-driven and are useful to extract a large number of “factors” from high-dimensional data.

• Population based, adults (Human Connectome Project)

• Children and adolescent population, enriched for mental distress (ABCD Cohort)

• Behavioural, demographic & environmental measures

• Self-report of (family) diagnosis and substance use

• Cognitive task performance

[5, 6, 39, 43]

Note: For consistency and clarity, throughout the paper the term “factors” is used to describe all kinds of factors, components, dimensions, sources, or modes, even if the term “factors” is unusual for the particular method that was used. For the purpose of the present paper, the interpretation is the same across these terms.

¹this references list is not exhaustive.