Package 'ProteinGymR'

AlphaMissense scores for ProteinGym variants

Description

AlphaMissense scores for ProteinGym variants

Usage

am_scores(metadata = FALSE)

Arguments

metadata

Logical, whether only experiment metadata should be returned. Default behavior is to return processed data with metadata included.

Details

am_scores() loads in the AlphaMissense pathogencity scores for substitutions matching those in the ProteinGym DMS assays. The table is taken from the AlphaMissense Supplementary Data by Cheng et al. 2023. See reference for details.

The columns contain:

DMS_id:

Character, ProteinGym assay identifier.

Uniprot_ID:

Character, UniProt accession identifier.

variant_id:

Character, variant identifier string matching ProteinGym. Protein position in the middle, and the reference and mutant amino acid residues to the left and right of the position, respectively.

AlphaMissense:

Numeric, AlphaMissense pathogenicity score.

Value

Returns a data.frame().

References

Cheng et al. (2023) Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 391, eadg7492. DOI:10.1126/science.adg7492.

Examples

data <- am_scores()
data_meta <- am_scores(metadata = TRUE)

---

Benchmark Variant Effect Prediction Models

Description

benchmark_models() plots one of the five model performance metrics ("AUC", "MCC", "NDCG", "Spearman", "Top_recall") for up to 5 user-specified variant effect prediction tools listed in available_models(). See reference for more details about the metrics and models.

Usage

available_models()

supervised_available_models()

benchmark_models(
  metric = c("AUC", "MCC", "NDCG", "Spearman", "Top_recall"),
  models = available_models()
)

Arguments

metric	character() a model performance metric to benchmark ("AUC", "MCC", "NDCG", "Spearman", "Top_recall").
models	character() a character vector of up to five variant effect prediction models to compare. Valid models can be seen with available_models().

Value

benchmark_models() returns a ggplot object visualizing a chosen model performance metric across several variant effect prediction models, ordered by highest to lowest mean performance score.

References

Notin, P., Kollasch, A., Ritter, D., van Niekerk, L., Paul, S., Spinner, H., Rollins, N., Shaw, A., Orenbuch, R., Weitzman, R., Frazer, J., Dias, M., Franceschi, D., Gal, Y., & Marks, D. (2023). ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design. In A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, & S. Levine (Eds.), Advances in Neural Information Processing Systems (Vol. 36, pp. 64331-64379). Curran Associates, Inc.

Examples

# Currently support models
available_models()

benchmark_models(metric = "Spearman", models = c("Site_Independent", 
"DeepSequence_single", "ESM2_15B", "GEMME", "CARP_640M"))

benchmark_models(models = "GEMME")

---

Integrate ProteinGym DMS and Model Prediction Scores

Description

dms_corr_plot() runs a Spearman correlation between ProteinGym deep mutational scanning (DMS) assay scores and predicted model scores. Returns a ggplot object for visualization.

Usage

dms_corr_plot(
  uniprotId,
  alphamissense_table = lifecycle::deprecated(),
  dms_table,
  model = "AlphaMissense"
)

Arguments

uniprotId	character() a valid UniProt accession identifier.
alphamissense_table	deprecated as of ProteinGymR 1.2.0. To access AlphaMissense model predictions, get model argument to "AlphaMissense".
dms_table	a table containing deep mutational scanning (DMS) assay scores for mutations. The default table loads substitutions from ProteinGym. Alternatively, a user-defined tibble::tbl_df or data.frame can be supplied.
model	character() a model to plot. To view the possible zero-shot and semi-supervised models available in ProteinGym v1.2 run ProteinGymR::available_models() or ProteinGymR::supervised_available_models(), respectively, or set ⁠model = "AlphaMissense⁠ to access AlphaMissense predictions. If no model argument is specified, the default loads in the supplemental table from the AlphaMissense paper.

Details

For dms_corr_plot(), model_table columns must include:

• UniProt_id: UniProt accession identifier.

• mutant: Mutant identifier string matching the dms_table format. Protein position in the middle, and the reference and mutant amino acid residues to the left and right of the position, respectively.

• {{model}}: Predicted model scores. Set this column name as the model name.

dms_table columns must include:

• UniProt_id: UniProt accession identifier.

• mutant: Mutant identifier string matching model_table variants. Specifically, the set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

• DMS_score: Experimental measurement in the DMS assay. Higher values indicate higher fitness of the mutated protein.

Value

dms_corr_plot() returns a ggplot object visualizing the Spearman correlation between experimental DMS scores and predicted model scores and prints the r and p-value of the analysis to console.

References

Cheng et al., Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 381, eadg7492. DOI:10.1126/science.adg7492.

Notin, P., Kollasch, A., Ritter, D., van Niekerk, L., Paul, S., Spinner, H., Rollins, N., Shaw, A., Orenbuch, R., Weitzman, R., Frazer, J., Dias, M., Franceschi, D., Gal, Y., & Marks, D. (2023). ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design. In A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, & S. Levine (Eds.), Advances in Neural Information Processing Systems (Vol. 36, pp. 64331-64379). Curran Associates, Inc.

Examples

# Use defaults. Only requires uniprotId
dms_corr_plot(uniprotId = "Q9NV35")

dms_corr_plot(
    uniprotId = "P04637",
    model = "Kermut"
)

---

ProteinGym Deep Mutational Scanning (DMS) Scores for Substitutions

Description

ProteinGym Deep Mutational Scanning (DMS) Scores for Substitutions

Usage

dms_substitutions(metadata = FALSE)

Arguments

metadata

Logical, whether only experiment metadata should be returned. Default behavior is to return processed data with metadata included.

Details

dms_substitutions() loads in ProteinGym deep mutational scanning assays (DMS) scores for substitutions in 217 studies. The data is provided by Notin et. al 2023. See reference for details.

Each assay includes 6 columns:

UniProt_id:

Character, UniProt accession identifier.

DMS_id:

Character, ProteinGym assay identifier.

mutant:

Character, set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

mutated_sequence:

Character, full amino acid sequence for the mutated protein.

DMS_score:

Numeric, experimental measurement in the DMS assay. Higher values indicate higher fitness of the mutated protein.

DMS_score_bin:

Factor, indicates whether the DMS_score is above the fitness cutoff (1 is fit, 0 is not fit).

Value

Returns a list() object of 217 individual assays.

References

Notin, P., Kollasch, A., Ritter, D., van Niekerk, L., Paul, S., Spinner, H., Rollins, N., Shaw, A., Orenbuch, R., Weitzman, R., Frazer, J., Dias, M., Franceschi, D., Gal, Y., & Marks, D. (2023). ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design. In A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, & S. Levine (Eds.), Advances in Neural Information Processing Systems (Vol. 36, pp. 64331-64379). Curran Associates, Inc.

Examples

data <- dms_substitutions()
data_meta <- dms_substitutions(metadata = TRUE)

---

Compare Prediction Scores for ProteinGym Models

Description

model_corr_plot() runs a Spearman correlation between predicted model scores for two models in ProteinGym v1.2. Returns a ggplot object for visualization.

Usage

model_corr_plot(uniprotId, model1 = "AlphaMissense", model2 = "GEMME")

Arguments

uniprotId	character() a valid UniProt accession identifier.
model1	character() first model to plot.
model2	character() second model to plot.

Details

For model_corr_plot():

model1 and model2 must be valid models. To view the possible zero-shot and semi-supervised models available in ProteinGym v1.2 run ProteinGymR::available_models() or ProteinGymR::supervised_available_models(), respectively, or set the model to 'AlphaMissense' for AlphaMissense predictions. If no models are specified, the default loads in AlphaMissense and GEMME.

Value

model_corr_plot() returns a ggplot object visualizing the Spearman correlation between the predicted scores generated by two models in ProteinGym and prints the p-value of the analysis to the console.

References

Cheng et al., Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 381, eadg7492. DOI:10.1126/science.adg7492.

Notin, P., Kollasch, A., Ritter, D., van Niekerk, L., Paul, S., Spinner, H., Rollins, N., Shaw, A., Orenbuch, R., Weitzman, R., Frazer, J., Dias, M., Franceschi, D., Gal, Y., & Marks, D. (2023). ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design. In A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, & S. Levine (Eds.), Advances in Neural Information Processing Systems (Vol. 36, pp. 64331-64379). Curran Associates, Inc.

Examples

# Use defaults. Only requires uniprotId
model_corr_plot(uniprotId = "Q9NV35")

model_corr_plot(
    uniprotId = "P04637",
    model1 = "Kermut",
    model2 = "EVE_single"
)

---

Visualize DMS Scores Along a Protein

Description

plot_dms_heatmap() plots DMS scores for amino acid substitutions along a protein in a defined DMS assay.

Usage

plot_dms_heatmap(
  assay_name,
  dms_data,
  start_pos = NULL,
  end_pos = NULL,
  exact_coord = FALSE,
  cluster_rows = FALSE,
  cluster_columns = FALSE,
  color_scheme,
  ...
)

Arguments

assay_name	character() a valid DMS assay name. For the full list of available assays, run names() on the list object loaded with ProteinGymR::dms_substitutions(). Alternatively, the name of a user-defined DMS assay.
dms_data	list() object of DMS assays loaded with ProteinGymR::dms_substitutions() if not specified by user. Alternatively, a user-defined list of DMS assays with names corresponding to assay_name param.
start_pos	integer() first amino acid position to plot. If missing, default start is at the first position along the protein where DMS scores are available.
end_pos	integer() last amino acid position to plot. If missing, default end is at the last position along the protein where DMS scores are available.
exact_coord	logical() TRUE will plot the precise start_pos and end_pos coordinates defined. By default, exact_coord is set to FALSE, plotting only amino acid positions with available data in the chosen assay.
cluster_rows	logical() defaults to FALSE. See argument details in ComplexHeatmap::Heatmap.
cluster_columns	logical() defaults to FALSE. See argument details in ComplexHeatmap::Heatmap.
color_scheme	character() defaults to blue, white, and red to represent positive, neutral, negative scores. Set argument equal to "EVE" to use the color scheme consistent with the popEVE portal.
...	additional arguments passed to internal plotting functions.

Details

For plot_dms_heatmap(), dms_data must be a list() object with set names for each assay element matching assay_name parameter.

Each assay in the dms_data() must include the following columns:

• mutant: Mutant identifier string matching. Specifically, the set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

• DMS_score: Experimental measurement in the DMS assay. Higher values indicate higher fitness of the mutated protein.

Value

Returns a ComplexHeatmap::Heatmap plot of DMS scores for each position along a protein in a chosen DMS assay. The x-axis shows amino acid positions where a DMS mutation exist, and the y-axis represents possible amino acid residues, ordered by default based on the physiochemical groupings. Higher and lower DMS scores indicate a more positive or negative fitness effect after the mutation, respectively.

Examples

dms_data <- dms_substitutions()

plot_dms_heatmap(assay_name = "A0A192B1T2_9HIV1_Haddox_2018", 
    start_pos = 10, 
    end_pos = 80)
    
plot_dms_heatmap(assay_name = "A0A192B1T2_9HIV1_Haddox_2018", 
    start_pos = 10, 
    end_pos = 80, 
    exact_coord = TRUE,
    color_scheme = "EVE")
    
plot_dms_heatmap(assay_name = "A0A192B1T2_9HIV1_Haddox_2018", 
    start_pos = 50, 
    end_pos = 100, 
    cluster_rows = TRUE)

---

Visualize DMS and Model Scores on 3D Protein Structures

Description

plot_structure() plots DMS or model scores for amino acid substitutions on a 3D protein structure for a chosen assay.

Usage

plot_structure(
  assay_name,
  pdb_file,
  data_scores = "DMS",
  dms_data = NULL,
  start_pos = NULL,
  end_pos = NULL,
  full_structure = FALSE,
  aggregate_fun = mean,
  color_scheme = NULL
)

Arguments

assay_name	character() a valid DMS assay name. For the full list of available assays, run names() on the list object loaded with ProteinGymR::dms_substitutions(). Alternatively, the name of a
pdb_file	string() defaults to corresonding PDB FilePath on ExperimentHub. Alternatively, a file path to a user-defined PDB file.
data_scores	character() specify whether DMS, zero-shot, or supervised model prediction scores should be displayed scores. Pass either "DMS" for experimental scores, or alternatively, a model name from available_models() for zero-shot or supervised_available_models() for semi-supervised models options. Defaults to DMS.
dms_data	list() object of DMS assays loaded with ProteinGymR::dms_substitutions(). Alternatively, a user-defined list of DMS assays with names corresponding to assay_name param.
start_pos	integer() first amino acid position to plot. If missing, default start is the first position along the protein in the PDB file.
end_pos	integer() last amino acid position to plot. If missing, default end is the last position along the protein in the PDB file.
full_structure	logical() defaults to FALSE and will only plot protein regions where there is DMS data available in the assay. If start_pos and end_pos coordinates are specified, plotting is restricted to this defined region. Setting full_structure() to TRUE will display full protein structure in the PBD file, and grey out regions where no DMS data is available.
aggregate_fun	method for aggregating DMS scores for each residue. For example, give min, max, or var to return the minimum, maximum, or variance of scores for each position, respectively. aggregate_fun can also take in a user-defined function with a numeric vector as input. By default, the mean DMS score across mutations at each position is calculated.
color_scheme	character() defaults to blue, white, and red to represent positive, neutral, negative scores. Set argument equal to "EVE" to use the color scheme consistent with the popEVE portal.

Details

By default, plot_structure() plots the mean DMS values of all amino acid residues, summarized for a protein position. If a model is chosen instead for data_scores argument, a helper function is invoked which normalizes the model prediction scores using a rank-based normal quantile transformation. The result is a set of normalized scores that preserve the rank order of the models scores, while standardizing the distribution. Transformed values typically fall between -3 and 3. This normalization ensures the scores are approximately standard normally distributed (mean = 0, SD = 1), allowing comparisons across models.

For plot_structure(), dms_data must be a list() object with set names for each assay element matching assay_name parameter.

Each assay in the dms_data() must include the following columns:

• mutant: Mutant identifier string matching. Specifically, the set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

• DMS_score: Experimental measurement in the DMS assay. Higher values indicate higher fitness of the mutated protein.

Each PBD table in pdb_file must include the following columns:

Value

plot_structure() returns a r3dmol::r3dmol object of DMS scores for each position along a protein in a chosen DMS assay. The x-axis shows amino acid positions where a DMS mutation exist, and the y-axis represents possible amino acid residues, ordered by default based on the physiochemical groupings. Higher and lower DMS scores indicate a more positive or negative fitness effect after the mutation, respectively.

Examples

plot_structure(assay_name = "C6KNH7_9INFA_Lee_2018",
   start_pos = 20, 
   end_pos = 50,
   full_structure = FALSE,
   aggregate_fun = max)
   
plot_structure(assay_name = "C6KNH7_9INFA_Lee_2018",
   start_pos = 20,
   end_pos = 50,
   data_scores = "GEMME")
   
plot_structure(assay_name = "ACE2_HUMAN_Chan_2020", 
    data_scores = "Kermut",
    color_scheme = "EVE")

---

Visualize Zero Shot Scores Along a Protein

Description

plot_zeroshot_heatmap() plots predicted model scores under the zero-shot model for amino acid substitutions along a protein in a defined DMS assay.

Usage

plot_zeroshot_heatmap(
  assay_name,
  model_data,
  model = NULL,
  start_pos = NULL,
  end_pos = NULL,
  exact_coord = FALSE,
  cluster_rows = FALSE,
  cluster_columns = FALSE,
  color_scheme,
  ...
)

Arguments

assay_name	character() a valid assay name. For the full list of available assays, run names() on the list object loaded with ProteinGymR::zeroshot_substitutions(). Alternatively, the name of a user-defined DMS assay.
model_data	list() object of zero-shot assays loaded with ProteinGymR::zeroshot_substitutions(). Alternatively, a user-defined list of assays with names corresponding to assay_name param.
model	character() one of the 79 zero-shot models to plot. To view the list of models, runs ProteinGymR::available_models().
start_pos	integer() first amino acid position to plot. If missing, default start is at the first position along the protein where zero shot scores are available.
end_pos	integer() last amino acid position to plot. If missing, default end is at the last position along the protein where zero shot scores are available.
exact_coord	logical() TRUE will plot the precise start_pos and end_pos coordinates defined. By default, exact_coord is set to FALSE, plotting only amino acid positions with available data in the chosen assay.
cluster_rows	logical() defaults to FALSE. See argument details in ComplexHeatmap::Heatmap.
cluster_columns	logical() defaults to FALSE. See argument details in ComplexHeatmap::Heatmap.
color_scheme	character() defaults to blue, white, and red to represent positive, neutral, negative scores. Set argument equal to "EVE" to use the color scheme consistent with the popEVE portal.
...	additional arguments passed to internal plotting functions.

Details

For plot_zeroshot_heatmap(), model_data must be a list() object with set names for each assay element matching assay_name parameter.

Each assay in the model_data() must include the following columns:

• mutant: Mutant identifier string matching. Specifically, the set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

• A column with predicted model scores for each substitution. Column name should match model argument string.

Value

Returns a ComplexHeatmap::Heatmap plot of zero-shot model scores for each position along a protein in a chosen DMS substitution assay. The x-axis shows amino acid positions where a DMS mutation exist, and the y-axis represents possible amino acid residues, ordered by default based on the physiochemical groupings. Higher and lower DMS scores indicate a more positive or negative fitness effect after the mutation, respectively.

Examples

available_models()

plot_zeroshot_heatmap(assay_name = "A0A192B1T2_9HIV1_Haddox_2018", 
    model = "GEMME",
    start_pos = 600,
    end_pos = 700, 
    color_scheme = "EVE")
    
plot_zeroshot_heatmap(assay_name = "SRC_HUMAN_Nguyen_2022",
    model = "CARP_38M")

---

Load Semi-Supervised Model Predictions for Substitutions in 217 Assays

Description

Load Semi-Supervised Model Predictions for Substitutions in 217 Assays

Load Semi-Supervised Model Summary Metrics

Usage

supervised_substitutions(
  metadata = FALSE,
  fold_scheme = c("contiguous", "modulo", "random")
)

supervised_metrics(metadata = FALSE)

Arguments

metadata	Logical, whether only experiment metadata should be returned. Default behavior is to return processed data with metadata included.
fold_scheme	Character, which validation folding scheme to load. Options include: "contiguous", "modulo", or "random". Default behavior loads "contiguous". For more information about the different folding schemes, refer to the original publication.

Details

supervised_substitutions() loads prediction scores outputted by semi-supervised models run on the 217 DMS substitution assays.

For raw model predictions, each assay includes 18 columns:

UniProt_id:

Character, UniProt accession identifier.

DMS_id:

Character, ProteinGym assay identifier.

mutant:

Character, set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

mutated_sequence:

Character, full amino acid sequence for the mutated protein.

DMS_score:

Numeric, experimental measurement in the DMS assay. Higher values indicate higher fitness of the mutated protein.

DMS_score_bin:

Factor, indicates whether the DMS_score is above the fitness cutoff (1 is fit, 0 is not fit).

Columns 7:18:

Respective semi-supervised model name.

supervised_metrics() loads in model performance summary metrics ("Spearman" and "MSE") from semi-supervised models in ProteinGymR run on 217 DMS substitution assays.

A metric summary table with 7 columns:

UniProt_id:

Character, UniProt accession identifier.

DMS_id:

Character, ProteinGym assay identifier.

mutant:

Character, set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

model_name:

Character, semi-supervised model used.

fold_variable_name:

Character, the folding scheme used.

Spearman:

Numeric, Spearman performance metric.

MSE:

Numeric, MSE the Spearman performance metric.

Value

Returns a list() object of 217 individual assays.

Returns a data.frame() with 7 columns.

References

Notin, P., Kollasch, A., Ritter, D., van Niekerk, L., Paul, S., Spinner, H., Rollins, N., Shaw, A., Orenbuch, R., Weitzman, R., Frazer, J., Dias, M., Franceschi, D., Gal, Y., & Marks, D. (2023). ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design. In A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, & S. Levine (Eds.), Advances in Neural Information Processing Systems (Vol. 36, pp. 64331-64379). Curran Associates, Inc.

Examples

data <- supervised_substitutions()
data_random <- supervised_substitutions(fold_scheme = "random")
meta <- supervised_substitutions(metadata = TRUE)

data <- supervised_metrics()
meta <- supervised_metrics(metadata = TRUE)

---

Load Zero-shot Model Predictions and Metrics for Substitutions in 217 DMS Assays

Description

Load Zero-shot Model Predictions and Metrics for Substitutions in 217 DMS Assays

Usage

zeroshot_DMS_metrics(metadata = FALSE)

zeroshot_substitutions(metadata = FALSE)

Arguments

metadata

Logical, whether only experiment metadata should be returned. Default behavior is to return processed data with metadata included.

Details

zeroshot_DMS_metrics() loads in the model performance metrics ("AUC", "MCC", "NDCG", "Spearman", "Top_recall") calculated on the DMS substitutions in the zero-shot setting for 79 models updated in ProteinGym v1.2.

Each data.frame contains the following columns:

DMS_ID:

Character, Assay name for the DMS study.

Columns 2:80:

Numeric, Corresponding to the average performance score of each of the 79 models tested.

Number_of_Mutants:

Numeric, Number of protein mutants evaluated.

Selection_Type:

Character, Protein function grouping.

UniProt_ID:

Character, UniProt protein entry name identifier.

DMS_score:

Numeric, experimental measurement in the DMS assay. Higher values indicate higher fitness of the mutated protein.

DMS_score_bin:

Factor, indicates whether the DMS_score is above the fitness cutoff (1 is fit, 0 is not fit).

MSA_Neff_L_category:

Multiple sequence alignment category.

Taxon:

Taxon group.

zeroshot_substitutions() loads prediction scores outputted by models in the zero-shot setting evaluated on the 217 DMS substitution assays. To examine all model options, run available_models().

For raw model predictions, each data.frame includes 85 columns:

UniProt_id:

Character, UniProt accession identifier.

DMS_id:

Character, ProteinGym assay identifier.

mutant:

Character, set of substitutions to apply on the reference sequence to obtain the mutated sequence (e.g., A1P:D2N implies the amino acid 'A' at position 1 should be replaced by 'P', and 'D' at position 2 should be replaced by 'N').

mutated_sequence:

Character, full amino acid sequence for the mutated protein.

DMS_score:

Numeric, experimental measurement in the DMS assay. Higher values indicate higher fitness of the mutated protein.

DMS_score_bin:

Factor, indicates whether the DMS_score is above the fitness cutoff (1 is fit, 0 is not fit).

Columns 7:85:

Respective zero-shot model name.

Value

Returns a list() object with 5 data.frame() corresponding to a model metric table.

Returns a list() object of 217 individual assays.

References

Notin, P., Kollasch, A., Ritter, D., van Niekerk, L., Paul, S., Spinner, H., Rollins, N., Shaw, A., Orenbuch, R., Weitzman, R., Frazer, J., Dias, M., Franceschi, D., Gal, Y., & Marks, D. (2023). ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design. In A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, & S. Levine (Eds.), Advances in Neural Information Processing Systems (Vol. 36, pp. 64331-64379). Curran Associates, Inc.

Examples

data <- zeroshot_DMS_metrics()
data_meta <- zeroshot_DMS_metrics(metadata = TRUE)

data <- zeroshot_substitutions()
data_meta <- zeroshot_substitutions(metadata = TRUE)

---