sync with latest AZ version

Author: Löffler, Hannes

CHANGELOG.md

@@ -10,6 +10,122 @@ This follows the guideline on [keep a changelog](https://keepachangelog.com/)
 - CAZP scoring component
 
 
+## [4.2.5] 2024-03-20
+
+### Added
+
+- Log network parameters
+
+
+## [4.2.4] 2024-03-20
+
+### Changed
+
+- Reworked TL code: clean-up, image layout, graph for Reinvent
+
+### Added
+
+- Added options and better statistics for TL: valid SMILES, duplicates
+- Disableable standardization
+
+### Removed
+
+- KL divergence in TB output
+
+### Fixed
+
+- Batch size calculation
+
+
+## [4.2.3] 2024-03-14
+
+### Fixed
+
+- Vocabulary for mol2mol and reinvent is saved as dictionary
+
+
+## [4.2.2] 2024-03-13
+
+### Fixed
+
+- Sigmoid functions in scoring have now a stable implementation
+
+
+## [4.2.1] 2024-03-12
+
+### Removed
+
+- Long chain (SMARTS for 5 aliphatic carbons) check
+
+
+## [4.2.0] 2024-03-08
+
+### Added
+
+- Unified transformer code to faciliate new model designs
+
+
+## [4.1.16] 2024-03-07
+
+### Fixed
+
+- filters now apply to transformed scores 
+
+
+## [4.1.15] 2024-03-07
+
+### Fixed
+
+- Minor change inception filter: cleaner way of handling internal SMILES store
+
+
+## [4.1.14] 2024-03-06
+
+### Added
+
+- Updated script for creating an empty classical Reinvent model
+
+### Fixed
+
+- Memory bug in TL related to similarity calculation: made this optional
+
+
+## [4.1.13] 2024-03-05
+
+### Added
+
+- Allowed runs with only filter/penalty components 
+
+
+## [4.1.12] 2024-03-05
+
+### Added
+
+- Better logging for Reinvent standardizer
+
+
+## [4.1.11] 2024-02-27
+
+### Fixed
+
+- Inception filters out SMILES containing tokens that are not compatible with the prior
+
+
+## [4.1.10] 2024-02-26
+
+### Fixed
+
+- Numerically stable double sigmoid implementation
+- Number of CPUs for TL (Mol2Mol) is now 1
+
+
+## [4.1.9] 2024-02-22
+
+### Fixed
+
+- Save models with no metadata
+
+
 ## [4.1.8] 2024-02-20
 
 ### Fixed

NEWS.md

@@ -1,7 +1,29 @@
+New in REINVENT 4.2
+===================
+
+For details see CHANGELOG.md.
+
+* Reworked TL code with added options and statistics
+* Standardization can be switched off in TL (useful in new prior creation)
+* Similarity calculation in TL made optional
+* Updated script for empty classical Reinvent model creation
+* Allow runs with only filter/penalty components
+* Stable sigmoid functions
+* Removed long chain check in SMILES processing
+* Unified transformer code
+* Filter apply to transformed scores
+* Better memory handling in inception
+* Better logging for Reinvent standardizer
+* Inception filters for tokens not compatible with the prior
+* Number of CPUs for TL (Mol2Mol pair generation) is 1 by default
+* Tensorboard histogram bug fixed again
+* Code improvements and fixes
+
+
 New in REINVENT 4.1
 ===================
 
-For details, see CHANGELOG.md.
+For details see CHANGELOG.md.
 
 * Scoring component MolVolume
 * Scoring component for all 210 RDKit descriptors

README.md

@@ -13,8 +13,9 @@ property profile defined as a multi-component score.  Transfer Learning (TL)
 can be used to create or pre-train a model that generates molecules closer
 to a set of input molecules. 
 
-A preprint with more details is available on ChemRxiv:
-[REINVENT4: Modern AI-Driven Generative Molecule Design](https://chemrxiv.org/engage/chemrxiv/article-details/65463cafc573f893f1cae33a).
+A paper describing the software has been published as Open Access in the
+Journal of Cheminformatics:
+[Reinvent 4: Modern AI–driven generative molecule design](https://link.springer.com/article/10.1186/s13321-024-00812-5?utm_source=rct_congratemailt&utm_medium=email&utm_campaign=oa_20240221&utm_content=10.1186/s13321-024-00812-5).
 See AUTHORS.md for references to previous papers.
 
 

configs/toml/PARAMS.md

@@ -59,8 +59,8 @@ Run transfer learning on a set of input SMILES.
 | num\_epochs            | number of epochs to run                                       |
 | save\_every\_n\_epochs | save checkpoint file every N epochs                           |
 | batch\_size            | batch size, note: affects SGD                                 |
-| num\_refs              | number of references for similarity                           |
-| sample\_batch\_size    | number of samples for similarity                              |
+| sample\_batch\_size    | batch size to calculate the sample loss and other statistics  |
+| num\_refs              | number of references for similarity if > 0, DO NOT use with large dataset (> 200 molecules) |
 | input\_model\_file     | filename of input prior model                                 |
 | smiles\_file           | SMILES file for Lib/Linkinvent and Molformer                  |
 | output_model\_file     | filename of the final model                                   |

configs/toml/transfer_learning.toml

@@ -16,7 +16,8 @@ num_epochs = 3  # number of steps to run
 save_every_n_epochs = 3  # save checkpoint model file very N steps
 batch_size = 50
 num_refs = 100  # number of reference molecules randomly chosen for similarity
-sample_batch_size = 100  # number of sampled molecules chosen for similarity
+                # set this to zero for large datasets (>200 molecules)!
+sample_batch_size = 100  # number of sampled molecules to compute sample loss
 # Uncomment one of the comment blocks below.  Each generator needs a model
 # file and possibly a SMILES file with seed structures.
 
@@ -44,12 +45,12 @@ pairs.max_cardinality = 199
 ### the molecule to be optimizable
 ## LibInvent
 #input_model_file = "priors/libinvent.prior"
-#smiles_file = "TL_libinvent_100.smi"  # read first 2 columns: warheads, linker
-#validation_smiles_file = "TL_libinvent_100.smi"  # read first 2 columns: warheads, linker
+#smiles_file = "TL_libinvent_100.smi"  # read first 2 columns: scaffold, R-groups
+#validation_smiles_file = "TL_libinvent_100.smi"  # read first 2 columns: scaffold, R-groups
 #output_model_file = "TL_libinvent.model"
 #
 ## LinkInvent
 #input_model_file = "priors/linkinvent.prior"
-#smiles_file = "TL_linkinvent_100.smi"  # read first 2 columns: scaffold, R-groups
+#smiles_file = "TL_linkinvent_100.smi"  # read first 2 columns: warheads, linker
 #output_model_file = "TL_linkinvent.model"
-#validation_smiles_file = "TL_linkinvent_100.smi"  # read first 2 columns: scaffold, R-groups
+#validation_smiles_file = "TL_linkinvent_100.smi"  # read first 2 columns: warheads, linker

reinvent/__init__.py

@@ -1 +1,42 @@
-from .logger import *
+"""REINVENT4
+
+Namespace setup and backward compatibility
+"""
+
+import sys
+
+from .logger import *
+from reinvent import models
+from reinvent.models.libinvent.models import vocabulary  # not sure why needed
+
+
+# support compatibility with older model files
+PATH_MAP = (
+    # R3
+    ("reinvent_models.reinvent_core", models.reinvent),
+    ("reinvent_models.reinvent_core.models", models.reinvent.models),
+    ("reinvent_models.reinvent_core.models.vocabulary", models.reinvent.models.vocabulary),
+    ("reinvent_models.lib_invent", models.libinvent),
+    ("reinvent_models.lib_invent.models", models.libinvent.models),
+    ("reinvent_models.lib_invent.models.vocabulary", models.libinvent.models.vocabulary),
+    ("reinvent_models.link_invent", models.linkinvent),
+    ("reinvent_models.link_invent.model_vocabulary", models.linkinvent.model_vocabulary),
+    (
+        "reinvent_models.link_invent.model_vocabulary.vocabulary",
+       models.linkinvent.model_vocabulary.vocabulary,
+    ),
+    (
+        "reinvent_models.link_invent.model_vocabulary.model_vocabulary",
+       models.linkinvent.model_vocabulary.model_vocabulary,
+    ),
+    (
+        "reinvent_models.link_invent.model_vocabulary.paired_model_vocabulary",
+       models.linkinvent.model_vocabulary.paired_model_vocabulary,
+    ),
+)
+
+for module, path in PATH_MAP:
+    if path is None:
+        sys.modules[module] = sys.modules[__name__]
+    else:
+        sys.modules[module] = path

reinvent/chemistry/enums/filter_types_enum.py

@@ -5,18 +5,17 @@ class FilterTypesEnum:
     REMOVE_HYDROGENS = "remove_hydrogens"
     REMOVE_SALTS = "remove_salts"
     GENERAL_CLEANUP = "general_cleanup"
-    TOKEN_FILTERS = "token_filters"
+    UNWANTED_PATTERNS = "unwanted_patterns"
     VOCABULARY_FILTER = "vocabulary_filter"
     VALID_SIZE = "valid_size"
     HEAVY_ATOM_FILTER = "heavy_atom_filter"
-    ALLOWED_ATOM_TYPE = "allowed_atom_type"
-    ALIPHATIC_CHAIN_FILTER = "aliphatic_chain_filter"
-
+    ALLOWED_ELEMENTS = "allowed_elements"
 
     # try to find the internal value and return
     def __getattr__(self, name):
         if name in self:
             return name
+
         raise AttributeError
 
     # prohibit any attempt to set any values