Partitioned analysis of selection.

Analysis inspired by Marina Escalera Zamudio marinaescalera@gmail.com

The purpose of this analysis is to compare selective pressures between partitions in an alignment of coding sequences. The input is comprised of a NEXUS file with at least two defined partitions (see data/Measles_Fusion_Complete.nex for an example) as CHARSET in the ASSUMPTION block.

The analysis fits a series of codon-substitution models (jointly to all partitions) and performs statistical comparisons between them to determine whether or not dN/dS distributions differ between partitions. The workflow is as follows, and is controlled by several options, described below.

1. Fit the GTR nucleotide model to obtain nucleotide substitution biases and (per partition) branch lengths. These estimates are used as initial values for subsequent optimizations. Nucleotide substitution biases and equilibirum frequency estimates are shared by all partitions.
2. (unless turned off by the --mg No option) Fit the Muse-Gaut 94 model with a single dN/dS ratio per partition. Report these estimates and corresponding confidence intervals.
   • Perform a likelihood ratio test for equality of partition dN/dS estimates. If the test is singificant, then mean selective pressures differ between partitions.
3. (unless turned off by the --bs-rel No option) Fit the Branch-site REL model with K rates (controlled by the --rates argument, default 2) to partitions. Each partition is therefore described by its own K-rate distribution of dN/dS values, which vary across both sites and branches
   • If any of the partitions have estimates of dN/dS > 1, perform the test for positive selection. This test compares the null of having all dN/dS values constrained to be ≤1, versus the alternative that dN/dS > 1 somewhere along the tree in at least one of the partitions. If the test is singificant, then there is evidence for positive selection on some of the partitions.
   • Compare the alternative model (different distributions on each of the partitions) to the null model, where a single K-rate dN/dS distribution is If the test is singificant, then there is evidence for differences in selective pressure between some of the partitions.

Important options

Branch subset

The test can be run on all branches in the tree or a subset of the branches in the tree (e.g. internal branches only). In the case where only some of the branches are selected, the "background" branches get their own (per partition) nuisance parameters (dN/dS) which are estimated but not used for testing. An important use-case is to test for selection on internal branches of pathogen trees to reduce dN/dS inflation issues due to "within-host" variation at the tips (see Pond et al and Pybus et al

Branch lengths

Branch lengths between partitions can be either compeletely independent or proporitional (default). The latter scenario assumes that relative lineage specific evolutionary rates (other than due to selective pressures) do not vary between partitions. The former scenario is more general, but has many more parameters, and assumes general heterotachy.

Invokation

This analysis has one required argument

• --alignment the alignment file (NEXUS with at least two partitions) and tree (either one shared by all partitions, or one per partition)

HyPhy will write Markdown output to the screen and a JSON file with detailed fit results. See example at the end of the document

Complete options list

Available analysis command line options
---------------------------------------
Use --option VALUE syntax to invoke
If a [reqired] option is not provided on the command line, the analysis will prompt for its value
[conditionally required] options may or not be required based on the values of other options

code
	Which genetic code should be used 
	defaut value: Universal

alignment [required]
	An in-frame codon alignment in one of the formats supported by HyPhy

branches 
	The set of branches to test [All, Internal, Leaves]
	defaut value: All

rates
	The number omega rate classes to include in the model [1-10, default 2]
	defaut value: partitioned.rate_classes [computed at run time]

branch-lengths
	Proportional branch lengths across partitions [Proportion or Unlinked]
	defaut value: Proportional

bs-rel
	Run BS-REL analyses [Yes, No]
	defaut value: Yes

mg
	Run MG94 analyses [Yes, No]
	defaut value: Yes

output
	Write the resulting JSON to this file (default is to save to the same path as the alignment file + 'partitioned.json')
	defaut value: partitioned.codon_data_info[terms.json.json] [computed at run time]

##Example run

hyphy PartitionedAnalysis.bf --alignment data/Measles_Fusion_Complete.nex --branches Internal

---

Analysis Description

Fit several models of partition-level dN/dS variation to partitioned data and perform inference on whether or not there is evidence of different selective pressures among partitions

• Requirements: in-frame codon alignment with at least two partitions and a phylogenetic tree

• Written by: Sergei L Kosakovsky Pond

• Contact Information: spond@temple.edu

• Analysis Version: 0.1

code –> Universal Loaded a multiple sequence alignment with 56 sequences, 550 codons, and 4 partitions from /Users/sergei/Development/hyphy-analyses/PartitionedSelection/data/Measles_Fusion_Complete.nex

branches –> Internal The number omega rate classes to include in the model (permissible range = [1,10], default value = 2, integer): rates –> 2

branch-lengths –> Proportional

bs-rel –> Yes

mg –> Yes

Obtaining branch lengths and nucleotide substitution biases under the nucleotide GTR model

• Log(L) = -4971.94, AIC-c = 10836.17 (444 estimated parameters)

Fitting MG94xREV with separate dN/dS ratios for each partition

• Log(L) = -4862.24, AIC-c = 9975.51 (125 estimated parameters)
• non-synonymous/synonymous rate ratio for G1|background = 0.2249
• non-synonymous/synonymous rate ratio for G1|test = 0.0838 (95% profile CI 0.0500- 0.1304)
• non-synonymous/synonymous rate ratio for G2|background = 0.1992
• non-synonymous/synonymous rate ratio for G2|test = 0.0351 (95% profile CI 0.0126- 0.0756)
• non-synonymous/synonymous rate ratio for G3|background = 0.3343
• non-synonymous/synonymous rate ratio for G3|test = 0.1020 (95% profile CI 0.0253- 0.2651)
• non-synonymous/synonymous rate ratio for G4|background = 0.1416
• non-synonymous/synonymous rate ratio for G4|test = 0.1203 (95% profile CI 0.0491- 0.2377)

Checking for difference in partition-level dN/dS using a likelihood ratio test

• Log(L) = -4864.73, AIC-c = 9974.44 (122 estimated parameters)
• Test for dN/dS equality across partitions: p-value = 0.1730, LRT = 4.98

Fitting partition-specific BS-REL models with 2 rate classes

• Log(L) = -4845.65, AIC-c = 10015.01 (161 estimated parameters)
• Inferred rate distribution for G1

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.001	98.788
Diversifying selection	9.314	1.212

• background rate distribution for G1

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	97.737
Diversifying selection	13.622	2.263

• Inferred rate distribution for G2

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.034	100.000
Diversifying selection	1.001	0.000	Not supported by data

• background rate distribution for G2

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	98.907
Diversifying selection	34.938	1.093

• Inferred rate distribution for G3

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.121	100.000
Diversifying selection	1.001	0.000	Not supported by data

• background rate distribution for G3

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	1.545
Negative selection	0.318	98.455

• Inferred rate distribution for G4

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.141	100.000
Diversifying selection	1.001	0.000	Not supported by data

• background rate distribution for G4

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.087	28.956
Negative selection	0.133	71.044

Testing for positive selection on any of the partitions

• Log(L) = -4846.39, AIC-c = 10008.41 (157 estimated parameters)
• Inferred rate distribution for G1

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	91.471
Neutral evolution	1.000	8.529

• background rate distribution for G1

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	97.749
Diversifying selection	13.820	2.251

• Inferred rate distribution for G2

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.034	100.000
Neutral evolution	1.000	0.000	Not supported by data

• background rate distribution for G2

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	98.957
Diversifying selection	37.388	1.043

• Inferred rate distribution for G3

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.121	100.000
Neutral evolution	1.000	0.000	Not supported by data

• background rate distribution for G3

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	6.244
Negative selection	0.333	93.756

• Inferred rate distribution for G4

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.141	100.000
Neutral evolution	1.000	0.000	Not supported by data

• background rate distribution for G4

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.093	24.374
Negative selection	0.128	75.626

• Test for positive selection on any of the partitions: p-value = 0.8299, LRT = 1.48

Testing for equality of rate distributions across partitions

• Log(L) = -4860.30, AIC-c = 10026.11 (152 estimated parameters)
• Test for selective differences between partitions: p-value = 0.0006, LRT = 29.28
• Inferred joint rate distribution

Selection mode	dN/dS	Proportion, %	Notes
Negative selection	0.000	98.437
Diversifying selection	11.486	1.563

---

Partitioned analyses test summary (corrected p-values)

• Likelihood ratio test for Test for selective differences between partitions, p = 0.0017.
• Likelihood ratio test for Test for dN/dS equality across partitions, p = 0.3460.
• Likelihood ratio test for Test for positive selection on any of the partitions, p = 0.8299.

---