update readme
Browse files- README.md +13 -43
- src/02.1_gather_ThermoMPNN_splits.py +0 -2
README.md
CHANGED
@@ -542,11 +542,8 @@ Of these
|
|
542 |
*** **IMPORTANT! Please [register your use](https://forms.gle/wuHv8MKmEu4EEMA99) of these data so that we (the Rocklin Lab) can continue to release new useful datasets!! This will take 10 seconds!!** ***
|
543 |
|
544 |
|
545 |
-
|
546 |
## Quickstart Usage
|
547 |
|
548 |
-
|
549 |
-
|
550 |
### Install HuggingFace Datasets package
|
551 |
|
552 |
Each subset can be loaded into python using the Huggingface [datasets](https://huggingface.co/docs/datasets/index) library.
|
@@ -565,7 +562,7 @@ then, from within python load the datasets library
|
|
565 |
|
566 |
### Load model datasets
|
567 |
|
568 |
-
To load one of the `MegaScale` model datasets, use `datasets.load_dataset(...)`:
|
569 |
|
570 |
>>> dataset_tag = "dataset3_single"
|
571 |
>>> dataset3_single = datasets.load_dataset(
|
@@ -643,13 +640,14 @@ Curated set of `325,132` ΔG measurements at `17,093` sites in `365` domains
|
|
643 |
|
644 |
**`dataset3_single`**:
|
645 |
The single point mutations in `dataset3`
|
646 |
-
* Using the train/val/test splits defined in ThermoMPNN
|
647 |
-
*
|
648 |
|
649 |
-
**`
|
650 |
The single point mutations in `dataset3`
|
651 |
-
* Using the 5-fold cross validation splits (`train_[0-4]`/`val_[0-4]`/`test_[0-4]`) defined in ThermoMPNN
|
652 |
|
|
|
|
|
653 |
|
654 |
### Target Selection
|
655 |
Targets consist of natural, designed, and destabilized wild-type
|
@@ -666,7 +664,7 @@ These were then processed by
|
|
666 |
which was used to trim amino acids from the N- and C termini that had a low number of contacts with any other residues.
|
667 |
* selected domains with up to 72 amino acids after excluding N- or C-terminal flexible loops
|
668 |
|
669 |
-
|
670 |
* previous Rosetta designs with ααα, αββα, βαββ, and ββαββ topologies (40 to 43 amino acids)
|
671 |
* new ββαα proteins designed using Rosetta (47 amino acids)
|
672 |
* new domains designed by trRosetta hallucination (46 to 69 amino acids)
|
@@ -715,37 +713,9 @@ Each target was analyzed and given a single quality category score G0-G11, which
|
|
715 |
* G11: Probably cleaved in folded state(s)
|
716 |
|
717 |
|
718 |
-
|
719 |
-
|
720 |
-
|
721 |
-
|
722 |
-
|
723 |
-
|
724 |
-
We determine ΔG using each sequence’s
|
725 |
-
* measured K50, a predicted sequence-specific K50 for the unfolded state (K50,U)
|
726 |
-
* a universal K50 for the folded state (K50,F)
|
727 |
-
|
728 |
-
|
729 |
-
published studies using purified protein samples for 1,188 variants of 10 proteins (Fig. 1g and Supplementary Fig. 1 for more details on GB129)
|
730 |
-
Our measurements for these sequences were all performed in libraries of 244,000–900,000 total sequences.
|
731 |
-
|
732 |
-
|
733 |
-
|
734 |
-
Other Datasets for comparison
|
735 |
-
* ProthermDB
|
736 |
-
* Thermodynamic data
|
737 |
-
* Thermal proteome profiling
|
738 |
-
* Rocklin2017
|
739 |
-
|
740 |
-
|
741 |
-
Tsuboyama2023_Dataset2_Dataset3_20230416.csv
|
742 |
-
* All sequences in dataset 2 and dataset 3 are included
|
743 |
-
* All sequences in this file have an inferred ΔG estimate
|
744 |
-
* only sequences in dataset 3 have a tabulated ΔΔG estimate
|
745 |
-
* datasets 2 and 3 include a very small number of sequences with low-quality data (wide confidence intervals)
|
746 |
-
because these sequences come from mutational scans that are high quality overall
|
747 |
-
* low-quality data (including mutant data filtered in Stage 3) have been filtered out and
|
748 |
-
replaced by a "–"" symbol in the columns labelled ‘_ML’ (for machine learning).
|
749 |
-
|
750 |
-
predicting wild-type amino acids from the folding stabilities (ΔG) of each protein variant
|
751 |
-
* 99,156 ΔG measurements (5,214 sites in 90 non-redundant natural domains)
|
|
|
542 |
*** **IMPORTANT! Please [register your use](https://forms.gle/wuHv8MKmEu4EEMA99) of these data so that we (the Rocklin Lab) can continue to release new useful datasets!! This will take 10 seconds!!** ***
|
543 |
|
544 |
|
|
|
545 |
## Quickstart Usage
|
546 |
|
|
|
|
|
547 |
### Install HuggingFace Datasets package
|
548 |
|
549 |
Each subset can be loaded into python using the Huggingface [datasets](https://huggingface.co/docs/datasets/index) library.
|
|
|
562 |
|
563 |
### Load model datasets
|
564 |
|
565 |
+
To load one of the `MegaScale` model datasets (see available datasets below), use `datasets.load_dataset(...)`:
|
566 |
|
567 |
>>> dataset_tag = "dataset3_single"
|
568 |
>>> dataset3_single = datasets.load_dataset(
|
|
|
640 |
|
641 |
**`dataset3_single`**:
|
642 |
The single point mutations in `dataset3`
|
643 |
+
* Using the train/val/test splits defined in ThermoMPNN [(Dieckhaus, et al., 2024)](https://www.pnas.org/doi/abs/10.1073/pnas.2314853121)
|
|
|
644 |
|
645 |
+
**`dataset3_single_cv`**:
|
646 |
The single point mutations in `dataset3`
|
647 |
+
* Using the 5-fold cross validation splits (`train_[0-4]`/`val_[0-4]`/`test_[0-4]`) defined in ThermoMPNN [(Dieckhaus, et al., 2024)](https://www.pnas.org/doi/abs/10.1073/pnas.2314853121)
|
648 |
|
649 |
+
**`AlphaFold_model_PDBs`**:
|
650 |
+
AlphaFold predicted structures of wildtype domains (even if structures exist in the Protein Databank)
|
651 |
|
652 |
### Target Selection
|
653 |
Targets consist of natural, designed, and destabilized wild-type
|
|
|
664 |
which was used to trim amino acids from the N- and C termini that had a low number of contacts with any other residues.
|
665 |
* selected domains with up to 72 amino acids after excluding N- or C-terminal flexible loops
|
666 |
|
667 |
+
**designed targets** were selected from
|
668 |
* previous Rosetta designs with ααα, αββα, βαββ, and ββαββ topologies (40 to 43 amino acids)
|
669 |
* new ββαα proteins designed using Rosetta (47 amino acids)
|
670 |
* new domains designed by trRosetta hallucination (46 to 69 amino acids)
|
|
|
713 |
* G11: Probably cleaved in folded state(s)
|
714 |
|
715 |
|
716 |
+
## ThermoMPNN splits
|
717 |
+
ThermoMPNN is a message passing neural network that predicts protein ΔΔG of mutation
|
718 |
+
based on ProteinMPNN [(Dauparas et al., 2022)](https://www.science.org/doi/10.1126/science.add2187).
|
719 |
+
ThermoMPNN uses in part data from the MegaScale dataset. From the mutations in `dataset2`,
|
720 |
+
272,712 mutations across 298 proteins were curated that were single point mutants, reliable,
|
721 |
+
and where the baseline is wildtype.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
src/02.1_gather_ThermoMPNN_splits.py
CHANGED
@@ -6,8 +6,6 @@ import pickle
|
|
6 |
|
7 |
|
8 |
|
9 |
-
|
10 |
-
|
11 |
# Gabe requested that the splits defined in ThermoMPNN of the MegaScale dataset be the default splits
|
12 |
|
13 |
# ThermoMPNN/datsets.py
|
|
|
6 |
|
7 |
|
8 |
|
|
|
|
|
9 |
# Gabe requested that the splits defined in ThermoMPNN of the MegaScale dataset be the default splits
|
10 |
|
11 |
# ThermoMPNN/datsets.py
|