This package provides a pipeline that allows for that training and evaluation of linear models for protein function prediction from user provided labelled AA sequences. This package also provides visualizations of the patterns learned by the linear models and provides functions that can be used to generate novel proteins with optimised functional characteristics.
# install.packages("devtools")
devtools::install_github("benz0id/linProt", build_vignettes = TRUE)
library("linProt")ls("package:linProt")
data(package = "linProt")
browseVignettes("linProt")This package works with continuously distributed protein functional characteristics, such as peak light absorption wavelength or fluorescence.
In the context of protein function prediction, linear models assign weights to each of the amino acid positions along an alignment. We can interpret these weights to gain insight into how residues contribute to the protein’s function and use the model to make predictions on unseen sequences.
This model takes a list of aligned amino acid sequences, each with a label denoting their experimentally determined function. These sequences are then encoded as either as A x 20 one hot matricies (where A is the number of residues in the aligned protein) or as A x 8 matricies, using the VHSE8 scores of each residue. After being partitioned into training and validation sets, these encoded data are then used to train a regularised linear model.
This trained linear model can be used for the function prediction of novel sequences, the design of maximally or minimally functional proteins, or to gain insight into the functional characteristics of the protein.
In order to achieve this, the package implements 8 functions that can be ordered form a modular pipeline. These include:
shuffled_partitions - Shuffle, partition, and encode the supplied peptide alignment using one of the following encoding functions
encode_onehot - Encode a peptide alignment as a tensor of one-hot vectors.
encode_physchem - Encode a peptide alignment as a tensor of vectors encoded by the VHSE matrix.
linear_train - Train and return a linear model that predicts protein function on using some training and validation data given some hyperparameters.
predictions - Use a linear model to make function predictions for some aligned peptide sequences.
maximal_sequence - Create a peptide that would have the maximum (or minimum) possible function as predicted by some model.
plot_cost_over_rep - Show the change in cost over the duration of some linear model’s training cycles. Helpful for discovering overfitting.
property_effect_heatmap - Using the weights supplied by a model trained on VHSE8 encoded peptides, show the learned effect of each VHSE on protein function.
residue_effect_heatmap - Using the weights supplied by a model trained on one hot encoded peptides, show the learned effects that having any given residue at any given position will have on protein function.
With this package, you can easily generate, evaluate, and use linear models for protein function/fitness predictionn tasks.
Here, we generate a linear model to peodict the absorption properties of various aligned channel rhodopin proteins using the provided dataset.
library(linProt)
data(rhoData)
examples <- rhoData$data
labels <- rhoData$labels
examples[1]
#> $`1`
#> [1] "--------------------------------------------------mfai----npeymn---------------------------et----vlld-------ec--tpiy-----ldigplweqvvar--vtqwfg-vilslvfliyyiwntykatcg-weelyvctvefckiiielyf-eytppam-------------------ifqtngqvtpwlryaewlltcpvilihls-nitgl--------ndd---ysgrtms-litsdlggicmavtaalsk-------g-----wlka-lffvigcg-ygastfynaaciyiesy----------------y------------------tm--p---qgicrrlvlwmagvfftswfmfpglflagpe-gtq---------al-swagttightvadllsknawgmighflrveihkhiiihgd-------vrrpvtvkalgrqvsvncfvdke------eeeederi--------"
labels[1]
#> [1] 436
# Shuffle, partition, and encode data set.
shuffled_datasets <- shuffled_partitions(examples, labels, num_p1 = 650,
encode=encode_onehot)
# Train a linear model to perform regression.
model <- linear_train(train_data = shuffled_datasets$e1,
train_labels = shuffled_datasets$l1,
valid_data = shuffled_datasets$e2,
valid_labels = shuffled_datasets$l2)
# View the expected influence of each residue on the function, (lambda max
# in this case).
residue_effect_heatmap(model, 380, 410)
plot_cost_over_rep(model)
# Make some predictions.
results <- predictions(shuffled_datasets$e2, model)
as.integer(results[1:5])
#> [1] 537 551 539 590 590
shuffled_datasets$l2[1:5]
#> [1] 516 526 566 566 604See the vignettes for a detailed description of the available functions, which can provide a manner of different visualizations of the model’s attributes, as well as apply it to protein engineering tasks.
This package was written by Benjamin Tudor Price. The R package
ggplot2 was used for visualization of gradient descent progress in the
plot_cost_over_rep function. The R package gplots was used to
visualise the learned weights using a heatmap for the
property_effect_heatmap and residue_effect_heatmap functions. The R
packages assertthat and assert were used to check user input
throughout the project. In the function linear_train, The R package
progess was used to produce an informative progress bar. The R package
stats was also used in linear_train in order to help format
variables. The shiny package was used to develop the shiny app.
The included dataset used algined sequence data published by (Karasuyama et al., 2018)
The VHSE matrix used (VHSE8) was originally released by (Mei et al., 2005)
Mei, H., Liao, Z. H., Zhou, Y., & Li, S. Z. (2005). A new set of amino acid descriptors and its application in peptide QSARs. Biopolymers, 80(6), 775–786. https://doi.org/10.1002/bip.20296
Xu, Y., Verma, D., Sheridan, R. P., Liaw, A., Ma, J., Marshall, N. M., McIntosh, J., Sherer, E. C., Svetnik, V., & Johnston, J. M. (2020). Deep Dive into Machine Learning Models for Protein Engineering. Journal of Chemical Information and Modeling, 60(6), 2773–2790. https://doi.org/10.1021/acs.jcim.0c00073
Karasuyama, M., Inoue, K., Nakamura, R., Kandori, H., & Takeuchi, I. (2018). Understanding Colour Tuning Rules and Predicting Absorption Wavelengths of Microbial Rhodopsins by Data-Driven Machine-Learning Approach. Scientific Reports, 8(1). https://doi.org/10.1038/s41598-018-33984-w
Bedbrook, C. N., Yang, K. K., Robinson, J. E., Mackey, E. D., Gradinaru, V., & Arnold, F. H. (2019). Machine learning-guided channelrhodopsin engineering enables minimally invasive optogenetics. Nature Methods, 16(11), 1176–1184. ://doi.org/10.1038/s41592-019-0583-8
H. Wickham. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York, 2016.
Warnes G, Bolker B, Bonebakker L, Gentleman R, Huber W, Liaw A, Lumley T, Maechler M, Magnusson A, Moeller S, Schwartz M, Venables B (2022). gplots: Various R Programming Tools for Plotting Data. R package version 3.1.3, https://CRAN.R-project.org/package=gplots.
Wickham H (2019). assertthat: Easy Pre and Post Assertions. R package version 0.2.1, https://CRAN.R-project.org/package=assertthat.
Binette O (2020). assert: Validate Function Arguments. R package version 1.0.1, https://CRAN.R-project.org/package=assert.
Csárdi G, FitzJohn R (2019). progress: Terminal Progress Bars. R package version 1.2.2, https://CRAN.R-project.org/package=progress.
R Core Team (2022). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
Yihui Xie (2022). knitr: A General-Purpose Package for Dynamic Report Generation in R. R package version 1.40.
This package was developed as part of an assessment for 2022 BCB410H:
Applied Bioinformatics course at the University of Toronto, Toronto,
CANADA. linProt welcomes issues, enhancement requests, and other
contributions. This package was improved by the advice of my peers and
Dr. Anjali Silva. To submit an issue, use the GitHub issues.