Add random_seed for AF2, Boltz and ColabFold

.github/workflows/workflow_json.yml

@@ -4,6 +4,8 @@ on:
   pull_request_target:
     branches:
       - dev
+    paths-ignore:
+      - "assets/workflow_jsons/**"
 
 concurrency:
   group: "${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}"
@@ -12,7 +14,6 @@ concurrency:
 jobs:
   generate-workflow-json:
     name: Create workflow JSON using bioflow-insight
-    if: github.actor != 'nf-core-bot'
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6

.nf-core.yml

@@ -4,7 +4,7 @@ lint:
     - .github/CONTRIBUTING.md
   actions_schema_validation: false
   multiqc_config: false
-nf_core_version: 3.5.1
+nf_core_version: 3.5.2
 repository_type: pipeline
 template:
   author: Athanasios Baltzis, Jose Espinosa-Carrasco, Harshil Patel

CHANGELOG.md

@@ -8,10 +8,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Enhancements & fixes
 
 - [[PR ##573](https://github.com/nf-core/proteinfold/pull/573)] - Adds affiliations for UNSW Structural Biology Facility (SBF).
+- [[PR #588](https://github.com/nf-core/proteinfold/pulls/588)] - Add `--random_seed` for AF2, Boltz, and ColabFold, removing and replacing `--alphafold2_random_seed`.
 
-| Old parameter | New parameter |
-| ------------- | ------------- |
-|               |               |
+| Old parameter              | New parameter   |
+| -------------------------- | --------------- |
+| `--alphafold2_random_seed` | `--random_seed` |
 
 > **NB:** Parameter has been **updated** if both old and new parameter information is present.
 > **NB:** Parameter has been **added** if just the new parameter information is present.

conf/modules_alphafold2.config

@@ -97,7 +97,7 @@ process {
         ext.args   = [
             params.use_gpu ? '--use_gpu_relax=true' : '--use_gpu_relax=false',
             params.alphafold2_max_template_date ? "--max_template_date ${params.alphafold2_max_template_date}" : '',
-            params.alphafold2_random_seed ? "--random_seed=${params.alphafold2_random_seed}" : ''
+            params.random_seed != null ? "--random_seed=${params.random_seed}" : ''
         ].join(' ').trim()
         publishDir = [
             [
@@ -156,7 +156,7 @@ process {
         }
         ext.args    = [
             params.use_gpu ? '--use_gpu_relax=true' : '--use_gpu_relax=false',
-            params.alphafold2_random_seed ? "--random_seed=${params.alphafold2_random_seed}" : ''
+            params.random_seed != null ? "--random_seed=${params.random_seed}" : ''
         ].join(' ').trim()
         publishDir = [
             [

conf/modules_boltz.config

@@ -69,6 +69,7 @@ process {
         }
         ext.args = [
             params.boltz_model          ? "--model ${params.boltz_model}"             : "",
+            params.random_seed != null  ? "--seed ${params.random_seed}"              : "",
             params.use_msa_server       ? "--use_msa_server"                          : "",
             params.msa_server_url       ? "--msa_server_url ${params.msa_server_url}" : "",
             params.boltz_use_potentials ? "--use_potentials"                          : "",

conf/modules_colabfold.config

@@ -40,11 +40,11 @@ process {
         if (params.use_gpu) {
             accelerator = 1
         }
-
         ext.args    = [
             params.colabfold_use_gpu_relax  ? '--use-gpu-relax' : '',
             params.colabfold_use_amber      ? '--amber' : '',
             params.colabfold_use_templates  ? '--templates' : '',
+            params.random_seed != null      ? "--random-seed ${params.random_seed}" : '',
             params.use_msa_server && params.msa_server_url ? "--host-url ${params.msa_server_url}" : ''
         ].join(' ').trim()
         publishDir = [

docs/usage.md

@@ -160,6 +160,18 @@ Alternatively, the required data layout for each of the individual modes is desc
 
 > Omitting the `--db` flag will allow the pipeline to download the reference data required to execute the selected modes.
 
+## Random Seeds
+
+The pipeline-wide `--random_seed` parameter can be used to set the same seed for `alphafold2`, `boltz`, and `colabfold` modes only.
+
+If no seed is explicitly provided, each mode keeps its native default behaviour:
+
+- `alphafold2`: AlphaFold2 generates a random seed internally.
+- `boltz`: Boltz leaves the seed unset (`--seed` is not passed).
+- `colabfold`: ColabFold defaults to `--random-seed 0`.
+
+For reproducible results between runs, it is recommended to set an explicit seed.
+
 ## Foldseek structural similarity search
 
 The pipeline can run an optional Foldseek `easy-search` step on the top-ranked structure for each input/mode combination.

docs/usage/alphafold2.md

@@ -131,7 +131,6 @@ See the [AlphaFold2](https://github.com/google-deepmind/alphafold) documentation
 | Parameter                        | Default                       | Description                                                                                                                                                                                                                                                                                                                                                                             |
 | -------------------------------- | ----------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | `--alphafold2_full_dbs`          | `false`                       | bfd is a large environmental sequence database used to identify homologs. small bfd is a redundancy recuced version of the bfd database which can reduce the execution time of homolog search but may reduce the depth of the resulting MSA in some cases. `--alphafold2_full_dbs` ensures that the full version of bfd is used for search.                                             |
-| `--alphafold2_random_seed`       | `null`                        | AlphaFold2 model inference is a stochastic process. Fixing a numerical random seed ensures that results are reproducible between runs                                                                                                                                                                                                                                                   |
 | `--alphafold2_max_template_date` | `2038-01-19`                  | Structural templates from the PDB are used as additional context when making predictions. Molecules with solved structures in the PDB can be trivially predicted by using these structures as inputs. When benchmarking model performance it can be useful to restrict the use of templates to those deposited before a fixed date to ensure solved structures do not bias predictions. |
 | `--alphafold2_params_prefix`     | `alphafold_params_2022-12-06` | Specify the alphafold2 params used for prediction                                                                                                                                                                                                                                                                                                                                       |
 | `--alphafold2_model_preset`      | `monomer_ptm`                 | Specify the alphafold2 monomer preset used for prediction.                                                                                                                                                                                                                                                                                                                              |

nextflow.config

@@ -20,14 +20,14 @@ params {
     use_msa_server              = false
     msa_server_url              = null
     uniref30_prefix             = null
+    random_seed                 = null
 
     // Alphafold2 parameters
     alphafold2_mode                 = 'split_msa_prediction' // {standard, split_msa_prediction}
     alphafold2_max_template_date    = '2038-01-19'
     alphafold2_full_dbs             = null // true full_dbs, false reduced_dbs
     alphafold2_model_preset         = 'monomer_ptm' // single-entry FASTA only: {monomer, monomer_casp14, monomer_ptm}
     alphafold2_db                   = null
-    alphafold2_random_seed          = null
 
     // Alphafold2 links
     alphafold2_bfd_link             = null

nextflow_schema.json

@@ -92,6 +92,10 @@
                     "description": "Use the cloud MSA server",
                     "icon": "fas folder-open"
                 },
+                "random_seed": {
+                    "type": "integer",
+                    "description": "Global random seed for modules that natively support explicit seeding: alphafold2, boltz, and colabfold."
+                },
                 "msa_server_url": {
                     "type": "string",
                     "description": "Specify your custom MMSeqs2 API server url",
@@ -134,10 +138,6 @@
                     "enum": ["monomer", "monomer_casp14", "monomer_ptm"],
                     "fa_icon": "fas fa-stream"
                 },
-                "alphafold2_random_seed": {
-                    "type": "integer",
-                    "description": "Random seed to control stochastic alphafold inference."
-                },
                 "alphafold2_params_prefix": {
                     "type": "string",
                     "default": "alphafold_params_2022-12-06",

subworkflows/local/utils_nfcore_proteinfold_pipeline/main.nf

@@ -94,6 +94,8 @@ ${colors.purple}  nf-core/proteinfold ${workflow.manifest.version}${colors.reset
         nextflow_cli_args
     )
 
+    validateInputParameters()
+
     //
     // Create channel from input file provided through input
     //
@@ -189,9 +191,15 @@ workflow PIPELINE_COMPLETION {
 // Check and validate pipeline parameters
 //
 def validateInputParameters() {
-    if (params.mode.toLowerCase().split(",").contains("alphafold3")) {
+    def requestedModes = params.mode.toLowerCase().split(",")*.trim()
+
+    if (requestedModes.contains("alphafold3")) {
         alphafold3Warn(log)
     }
+
+    if (params.random_seed != null) {
+        randomSeedModeWarn(log, requestedModes)
+    }
 }
 
 def modeChannel(ch, mode) {
@@ -310,3 +318,15 @@ def alphafold3Warn(log) {
         "  More information here: \"https://github.com/google-deepmind/alphafold3/blob/main/README.md#alphafold-3-source-code-and-model-parameters.\"\n" +
         "==================================================================================="
 }
+
+def randomSeedModeWarn(log, requestedModes) {
+    def supportedModes = ['alphafold2', 'boltz', 'colabfold'] as Set
+    def unsupportedModes = requestedModes.findAll { !supportedModes.contains(it) }.unique().sort()
+
+    if (!unsupportedModes.isEmpty()) {
+        log.warn "=============================================================================\n" +
+            "  The global --random_seed parameter only applies to alphafold2, boltz, and colabfold.\n" +
+            "  It will be ignored for the following selected mode(s): ${unsupportedModes.join(', ')}.\n" +
+            "==================================================================================="
+    }
+}