rc-foundry 0.1.4__py3-none-any.whl → 0.1.6__py3-none-any.whl

rfd3/configs/datasets/val/unconditional.yaml

@@ -0,0 +1,9 @@
+defaults:
+  - design_validation_base
+  - _self_
+
+dataset:
+  data: ${paths.data.design_benchmark_data_dir}/monomer.json
+  subset_to_keys: null  # Specific keys in json to keep, ignores all others.
+  name: unconditional-design
+  eval_every_n: 1

rfd3/configs/datasets/val/unconditional_deep.yaml

@@ -0,0 +1,9 @@
+
+defaults:
+  - unconditional
+  - _self_
+
+dataset:
+  data: ${paths.data.design_benchmark_data_dir}/unconditional_deep.json
+  name: unconditional-design-deep
+  eval_every_n: 8

rfd3/configs/datasets/val/unindexed.yaml

@@ -0,0 +1,8 @@
+
+defaults:
+  - unconditional
+  - _self_
+
+dataset:
+  data: ${paths.data.design_benchmark_data_dir}/unindexed.json
+  name: unindexed-design

rfd3/configs/datasets/val/val_examples/bcov_ppi_easy_medium_with_ori.yaml

@@ -0,0 +1,151 @@
+global_args:
+  infer_ori_strategy: hotspots
+
+insulinr:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/insulin_target.pdb
+  contig: 100-100,/0,B1-150
+  contig_atoms: '{}'
+  length: 250-250
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B59:
+      CG,CZ: 1
+    B83:
+      CG,CZ: 1
+    B91:
+      CG,CZ: 1
+pdl1:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/5o45_pdl1.pdb
+  contig: 100-100,/0,B1-115
+  contig_atoms: '{}'
+  length: 215-215
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B40:
+      CG,CZ: 1
+    B99:
+      CG,SD: 1
+    B107:
+      CG,CZ: 1
+vegfr:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/vegfr_2x1w_and_af2_B.pdb
+  contig: 100-100,/0,B1-200
+  contig_atoms: '{}'
+  length: 300-300
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B13:
+      CG1,CG2: 1
+    B15:
+      CG,CZ: 1
+    B43:
+      CG,CZ: 1
+    B75:
+      CG,SD: 1
+    B89:
+      CD1,CG2: 1
+    B91:
+      CG1,CG2: 1
+    B187:
+      CG,CD1: 1
+rbd:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/COVID19_target.pdb
+  contig: 100-100,/0,B1-195
+  contig_atoms: '{}'
+  length: 295-295
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B89:
+      CG,CZ: 1
+    B121:
+      CG,CZ: 1
+    B123:
+      CG,CD1: 1
+    B124:
+      CG,CZ: 1
+    B141:
+      CG,CZ: 1
+    B157:
+      CG,CZ: 1
+    B163:
+      CG,CZ: 1
+    B165:
+      CG,CZ: 1
+    B173:
+      CG,CZ: 1
+cd28:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/cd28_1yjd_B.pdb
+  contig: 100-100,/0,B1-118
+  contig_atoms: '{}'
+  length: 218-218
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B51:
+      CG,CZ: 1
+    B61:
+      CG,CZ: 1
+    B99:
+      CG,SD: 1
+    B104:
+      CG,CZ: 1
+il2ra:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/il2ra_1z92_B.pdb
+  contig: 100-100,/0,B1-122
+  contig_atoms: '{}'
+  length: 222-222
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B3:
+      CG,CD1: 1
+    B26:
+      CG,SD: 1
+    B43:
+      CG,CD1: 1
+    B44:
+      CG,CZ: 1
+    B46:
+      CG,CD1: 1
+il10ra:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/il10rb_1lqs_B.pdb
+  contig: 100-100,/0,B1-207
+  contig_atoms: '{}'
+  length: 307-307
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B39:
+      CG,CD1: 1
+    B50:
+      CD1,CG2: 1
+    B59:
+      CG,CZ: 1
+    B63:
+      CA,CB: 1
+    B64:
+      CG1,CG2: 1
+    B66:
+      CG,CD1: 1
+tie2:
+  input: /projects/ml/aa_design/benchmarks/bcov_af3_ppi_benchmark/tie2_2gy5_official_B.pdb
+  contig: 100-100,/0,B1-188
+  contig_atoms: '{}'
+  length: 288-288
+  redesign_motif_sidechains: false
+  atom_level_hotspots:
+    B132:
+      CG1,CG2: 1
+    B134:
+      CG,CZ: 1
+    B135:
+      CG,CD: 1
+    B139:
+      CG,CZ: 1
+    B140:
+      CD1,CG2: 1
+    B154:
+      CG1,CG2: 1
+    B156:
+      CG,CD1: 1
+    B167:
+      CG,CZ: 1
+    B172:
+      CD1,CG2: 1

rfd3/configs/datasets/val/val_examples/bcov_ppi_easy_medium_with_ori_spoof_helical_bundle.yaml

@@ -0,0 +1,7 @@
+defaults:
+  - bcov_ppi_easy_medium_with_ori
+  - _self_
+
+global_args:
+  use_ss_conditioning: true
+  spoof_helical_bundle_ss_conditioning: true

rfd3/configs/datasets/val/val_examples/bcov_ppi_easy_medium_with_ori_varying_lengths.yaml

@@ -0,0 +1,28 @@
+defaults:
+  - bcov_ppi_easy_medium_with_ori
+  - _self_
+
+insulinr:
+  contig: 65-120,/0,B1-150
+  length: 215-270
+pdl1:
+  contig: 65-120,/0,B1-115
+  length: 215-215
+vegfr:
+  contig: 65-120,/0,B1-200
+  length: 300-300
+rbd:
+  contig: 65-120,/0,B1-195
+  length: 295-295
+cd28:
+  contig: 65-120,/0,B1-118
+  length: 218-218
+il2ra:
+  contig: 65-120,/0,B1-122
+  length: 222-222
+il10ra:
+  contig: 65-120,/0,B1-207
+  length: 307-307
+tie2:
+  contig: 65-120,/0,B1-188
+  length: 288-288

rfd3/configs/datasets/val/val_examples/bpem_ori_hb.yaml

@@ -0,0 +1,212 @@
+defaults:
+  - bcov_ppi_easy_medium_with_ori
+  - _self_
+
+# Note: I have removed the more "internal" hotspot atoms on residues that also have
+# hbonds to avoid giving conflicting information with the h-bond conditioning.
+# For targets with no polar hotspots, I have inserted h-bond conditioning to nearby
+# residues.
+
+insulinr:
+  hbond_donors:
+    B113:
+      NH1,NH2: 1
+    B116:
+      NZ: 1
+    B139:
+      NE2: 1
+  hbond_acceptors:
+    B115:
+      OE2: 1 # OE1 is already h-bonded
+    B139:
+      ND1: 1
+
+  atom_level_hotspots:
+    B59:
+      CG,CZ: 1
+    B83:
+      CG,CZ: 1
+    B91:
+      CG,CZ: 1
+pdl1:
+  hbond_donors:
+    B107:
+      OH: 1
+    B109:
+      NH1,NH2: 1
+  hbond_acceptors:
+    B40:
+      OH: 1
+    B107:
+      OH: 1
+  atom_level_hotspots:
+    B40:
+      CZ: 1
+    B99:
+      CG,SD: 1
+    B107:
+      CZ: 1
+vegfr:
+  hbond_donors:
+    B15:
+      OH: 1
+    B43:
+      OH: 1
+  hbond_acceptors:
+    B9: # ASP
+      OD1,OD2: 1
+    B15:
+      OH: 1
+    B43:
+      OH: 1
+  atom_level_hotspots:
+    B13:
+      CG1,CG2: 1
+    B15:
+      CZ: 1
+    B43:
+      CZ: 1
+    B75:
+      CG,SD: 1
+    B89:
+      CD1,CG2: 1
+    B91:
+      CG1,CG2: 1
+    B187:
+      CG,CD1: 1
+rbd:
+  hbond_donors:
+    B89:
+      OH: 1
+    B121:
+      OH: 1
+    B141:
+      OH: 1
+  hbond_acceptors:
+    B157:
+      OH: 1
+    B163:
+      OH: 1
+    B173:
+      OH: 1
+  atom_level_hotspots:
+    B89:
+      CZ: 1
+    B121:
+      CZ: 1
+    B123:
+      CG,CD1: 1
+    B124:
+      CG,CZ: 1
+    B141:
+      CZ: 1
+    B157:
+      CZ: 1
+    B163:
+      CZ: 1
+    B165:
+      CG,CZ: 1
+    B173:
+      CZ: 1
+cd28:
+  hbond_donors:
+    B51:
+      OH: 1
+    B61:
+      OH: 1
+    B104:
+      OH: 1
+  hbond_acceptors:
+    B51:
+      OH: 1
+    B61:
+      OH: 1
+    B104:
+      OH: 1
+  atom_level_hotspots:
+    B51:
+      CZ: 1
+    B61:
+      CZ: 1
+    B99:
+      CG,SD: 1
+    B104:
+      CZ: 1
+il2ra:
+  hbond_donors:
+    B44:
+      OH: 1
+  hbond_acceptors:
+    B44:
+      OH: 1
+    B77: # HIS
+      ND1: 1
+    B2: # GLU
+      OE1,OE2: 1
+  atom_level_hotspots:
+    B3:
+      CG,CD1: 1
+    B26:
+      CG,SD: 1
+    B43:
+      CG,CD1: 1
+    B44:
+      CZ: 1
+    B46:
+      CG,CD1: 1
+il10ra:
+  hbond_donors:
+    B48: # ASN
+      ND2: 1
+    B59:
+      OH: 1
+  hbond_acceptors:
+    B48: # ASN
+      OD1: 1
+    B59:
+      OH: 1
+  atom_level_hotspots:
+    B39:
+      CG,CD1: 1
+    B50:
+      CD1,CG2: 1
+    B59:
+      CZ: 1
+    B63:
+      CA,CB: 1
+    B64:
+      CG1,CG2: 1
+    B66:
+      CG,CD1: 1
+tie2:
+  hbond_donors:
+    B134:
+      OH: 1
+    B135:
+      NZ: 1
+    B167:
+      OH: 1
+  hbond_acceptors:
+    B134:
+      OH: 1
+    B167:
+      OH: 1
+  atom_level_hotspots:
+    B132:
+      CG1,CG2: 1
+    B134:
+      CZ: 1
+    B135:
+      CD: 1
+    B139:
+      CG,CZ: 1
+    B140:
+      CD1,CG2: 1
+    B154:
+      CG1,CG2: 1
+    B156:
+      CG,CD1: 1
+    B167:
+      CZ: 1
+    B172:
+      CD1,CG2: 1

rfd3/configs/debug/default.yaml

@@ -0,0 +1,64 @@
+# @package _global_
+
+defaults:
+  - override /logger: null
+
+# default debugging setup, runs 1 full epoch
+# other debugging configs can inherit from this one
+
+# overwrite task name so debugging logs are stored in separate folder
+task_name: "debug"
+
+extras:
+  ignore_warnings: False
+  enforce_tags: False
+
+# sets level of all command line loggers to 'DEBUG'
+# https://hydra.cc/docs/tutorials/basic/running_your_app/logging/
+hydra:
+  job_logging:
+    root:
+      level: DEBUG
+  # use the below to also set hydra loggers to 'DEBUG'
+  verbose: True
+
+# Print example ID before forward pass
+callbacks:
+  print_example_id_before_forward_pass:
+      _target_: foundry.callbacks.train_logging.PrintExampleIDBeforeForwardPassCallback
+
+dataloader:
+  train:
+    dataloader_params: 
+      batch_size: 1
+      num_workers: 0  # debuggers don't like multiprocessing -- work on main thread
+      pin_memory: False  # disable gpu memory pin
+      prefetch_factor: null # must be null for num_workers=0
+    n_fallback_retries: 0 # disable fallback retries for debugging
+
+  val:
+    dataloader_params:
+      batch_size: 1
+      num_workers: 0
+      pin_memory: False
+      prefetch_factor: null # must be null for num_workers=0
+
+
+datasets:
+  crop_size: 100  # set small crop size for quick debugging
+  diffusion_batch_size_train: 1
+  diffusion_batch_size_inference: 1
+  n_recycles_train: 1
+  n_recycles_validation: 1
+  n_msa: 128
+  key_to_balance: null # otherwise big examples will be processed first
+
+trainer:
+  devices_per_node: 1
+  limit_train_batches: 1
+  limit_val_batches: 1
+  validate_every_n_epochs: 1
+
+# Set tags to help identify debugging runs
+tags:
+  - debug

rfd3/configs/debug/train_specific_examples.yaml

@@ -0,0 +1,21 @@
+# @package _global_
+
+# See: https://hydra.cc/docs/patterns/configuring_experiments/
+
+# to execute this experiment run:
+# python train.py +debug=train_single_example [any other arguments] 
+
+defaults:
+  - default
+  - gpu
+
+datasets: 
+  # you can add specific example IDs here to load a subset of the dataset (training)
+  subset_to_example_ids:
+    - "{['pdb', 'pn_units']}{3px1}{1}{['A_3']}"
+  val: null
+
+tags:
+  - debug
+  - train
+  - specific-examples

rfd3/configs/dev.yaml

@@ -0,0 +1,9 @@
+# @package _global_
+# Inference engine config for development
+hydra:
+  searchpath:
+    - pkg://configs
+
+defaults:
+  - inference_engine: dev
+  - _self_

rfd3/configs/experiment/debug.yaml

@@ -0,0 +1,14 @@
+# @package _global_
+
+defaults:
+  - /debug/default
+  - override /logger: csv
+
+name: a14-debug
+tags:
+  - debug
+project: test
+ckpt_path: null
+
+trainer:
+  prevalidate: True

rfd3/configs/experiment/pretrain.yaml

@@ -0,0 +1,31 @@
+# @package _global_
+# Training configuration for RFD3
+
+name: train-base
+tags: [print-model]
+ckpt_path: null
+
+datasets:
+  diffusion_batch_size_train: 16
+  crop_size: 256
+  max_atoms_in_crop: 2560  # ~10x crop size.
+  global_transform_args:
+    train_conditions:
+      unconditional:
+        frequency: 2.0
+      island:
+        frequency: 2.0
+      sequence_design:
+        frequency: 0.5
+      tipatom:
+        frequency: 5.0
+      ppi:
+        frequency: 0.0
+  train:
+    # These are the ratios used in the preprint but we set all pdb sampling by default since not everyone might download the distillation data.
+    #pdb:
+      #probability: 0.10
+    #monomer_distillation:
+      #probability: 0.90
+    pdb:
+      probability: 1.0

rfd3/configs/experiment/test-uncond.yaml

@@ -0,0 +1,10 @@
+# @package _global_
+
+defaults:
+  - _self_
+
+datasets:
+  global_transform_args:
+    train_conditions:
+      unconditional:
+        frequency: 9999999999

rfd3/configs/experiment/test-unindexed.yaml

@@ -0,0 +1,21 @@
+# @package _global_
+
+defaults:
+  - _self_
+
+name: train-unindexed-main
+
+datasets:
+  diffusion_batch_size_train: 24
+  global_transform_args:
+    train_conditions:
+      unconditional:
+        frequency: 3.0
+      sequence_design:
+        frequency: 0.5
+      island:
+        frequency: 3.0
+        p_unindex_motif_tokens: 1.0
+      tipatom:
+        frequency: 3.0
+        p_unindex_motif_tokens: 1.0

rfd3/configs/hydra/default.yaml

@@ -0,0 +1,18 @@
+# https://hydra.cc/docs/configure_hydra/intro/
+
+# enable color logging (requires `colorlog` to be installed)
+# defaults:
+#   - override hydra_logging: colorlog
+#   - override job_logging: colorlog
+
+
+# output directory, generated dynamically on each run
+run:
+  dir: ${paths.log_dir}/${task_name}/${name}/${now:%Y-%m-%d}_${now:%H-%M}_JOB_${oc.env:SLURM_JOB_ID,default}
+
+# ... this is where the log file is written (i.e. the programs output)
+job_logging:
+  handlers:
+    file:
+      # Incorporates fix from https://github.com/facebookresearch/hydra/pull/2242
+      filename: ${hydra.runtime.output_dir}/experiment.log

rfd3/configs/hydra/no_logging.yaml

@@ -0,0 +1,7 @@
+defaults:
+  - override job_logging: disabled
+  - override hydra_logging: disabled
+
+output_subdir: null  
+run:  
+  dir: .

rfd3/configs/inference.yaml

@@ -0,0 +1,9 @@
+# @package _global_
+
+hydra:
+  searchpath:
+    - pkg://configs
+
+defaults:
+  - inference_engine: rfdiffusion3
+  - _self_

rfd3/configs/inference_engine/base.yaml

@@ -0,0 +1,15 @@
+# @package _global_
+
+defaults:
+  - /hydra: no_logging
+
+# Parameters for RFD3InferenceEngine.__init__()
+ckpt_path: ???
+num_nodes: 1
+devices_per_node: 1
+verbose: false
+seed: null
+
+# Parameters for RFD3InferenceEngine.run()
+inputs: ???
+out_dir: ???

rfd3/configs/inference_engine/dev.yaml

@@ -0,0 +1,20 @@
+# @package _global_
+defaults:
+  - rfdiffusion3
+  - _self_
+
+diffusion_batch_size: 8
+n_batches: 1
+seed: 42
+
+dump_trajectories: True
+verbose: true
+skip_existing: False
+cleanup_guideposts: False
+cleanup_virtual_atoms: False
+output_full_json: True
+
+inference_sampler:
+  gamma_0: 0.0
+
+out_dir: ./logs/benchmark