hcwang and qxdu edited on Aug 4, 2023, 1 version

The Gpro package allows the biologist to discover novel functinal promoters in undiscovered sequence space, and to perform a wide variety of other quality-control analyses.

The Gpro package mainly provides new sequence discovery algorithms using both Wasserstein Generative Adversarial Networks(WGAN) and Multinomial Diffusion Networks(Diffusion), which have complementary strengths. This package also provides promoter strength prediction algorithms using existed models from various papers, supporting the assesment of both prokaryote and eukaryotic, and covering a range of currently popular models, like DenseNet, DenstNet + LSTM(DenseLSTM), CNN(CNN_K15, CNN_Wangye), Multihead Attention + BiLSTM(AtnnBilstm). These models support the discovery of promoter space. In addition, the Gpro Suite also provides Optimizers for searching sequences for optimization in expression level. Finally, You can also control the sequence quality and visually analyze model performance with different Evaluators.

The Gpro package is comprised of a collection of tools that work together, as shown below. To see what has changed recently you can peruse the release notes. During the training process of the predictor, 70% is used as the training set by default, and 30% is used as the validation set.

Apart from executing the process from initiation to completion, each module mentioned above has been individually packaged and can be imported individually.

Credits: all the evaluators refer to codes of BioAutoMATED([10]), an AutoML pipeline for biological samples.We have fine-tuned these codes to make them more suitable for our tasks.

4.1 Generator

Generator	Description	Import Method	Class Name	Citations
WGAN	Wasserstein Generative Adversarial Networks	gpro.generator.wgan.wgan	WGAN_language	[1]
Diffusion	Multinomial Diffusion	gpro.generator.diffusion.diffusion	Diffusion_language	[2]
cGAN	conditional Generative Adversarial Networks	gpro.generator.others.cgan.cgan	DeepSeed	[6]
VAE	Variational Auto-Encoder	gpro.generator.others.vae.vae	SimpleVAE	[7]

4.2 Predictor

Predictor	Description	Import Method	Class Name	Citations
CNN_K15	CNN network for k1.5 virus promoter	gpro.predictor.cnn_k15.cnn_k15	CNN_K15_language
CNN_Wangye	CNN network proposed by Wangye	gpro.predictor.cnn_wangye.cnn_wangye	WangYeModel_language	[1]
DenseNet	Predictor based on DenseNet	gpro.predictor.densenet.densenet	DenseNet_language	[3]
DenseLSTM	Predictor based on DenseNet and LSTM	gpro.predictor.denselstm.denselstm	DenseLSTM_language	[3]
AttnBiLSTM	Predictor based on Multihead Attention Layer and Bi-directional LSTM	gpro.predictor.attnbilstm.attnbilstm	AttnBilstm_language	[3]
GRUClassifier	Binary Classifier based on GRU structure	gpro.predictor.others.GRUClassifier	GRUClassifier_language	[8]
DeepSTARR2	Regressive DeepSTARR2 predictor and corresponding Binary Classifier	gpro.predictor.deepstarr2.deepstarr2, gpro.predictor.deepstarr2.deepstarr2_binary	DeepSTARR2_language, DeepSTARR2_binary_language	[11], [12]

4.3 Optimizer

Optimizer	Description	Import Method	Class Name	Citations
Filter	Using predictive model for filtering after generating a large quantity of novel sequences	gpro.optimizer.model_driven.filter	Filter
Genetic	Using genetic algorithm for searching novel sequences that maximum the predictor	gpro.optimizer.heuristic.genetic	GeneticAlgorithm
Annealing	Using simulated annealing algorithm for searching novel sequences that maximum the predictor	gpro.optimizer.heuristic.annealing	AnnealingAlgorithm
GDS	Using gradient descending algorithm for searching best hidden space that maximum the predictor	gpro.optimizer.model_driven.gradient	GradientAlgorithm
Feedback	Using feedback strategy for directly searching training set that maximum the predictor	gpro.optimizer.model_driven.feedback	Feedback	[8]
Drift	Using random drift for simulating natural selection	gpro.optimizer.evolution.drift	Drift	[9]
SSWM	Using Strong-Selection-Weak Mutation for simulating natural selection	gpro.optimizer.evolution.sswm	SSWM	[9]

4.4 Evaluator

Evaluator	Description	Import Method	Function Name	Citations
Kmer frequency	Estimating the generated sequences similarity with natural sequences, a criteria for generator	gpro.evaluator.kmer	plot_kmer_with_model, plot_kmer	[1]
saturation mutagenesis	Mutate all site at the sequence set, using predictor for weighting the importance at each site	gpro.evaluator.mutagenesis	plot_mutagenesis
coefficient	Estimating the predicted expression levels' similarity with natural expressions, a criteria for predictor	gpro.evaluator.regression	plot_regression_performance	[5]
saliency map	Evaluate the importance of each location for the predictor	gpro.evaluator.saliency	plot_saliency_map
seqlogo	Plot the seqlogo that consistent with the predictor	gpro.evaluator.seqlogo	plot_seqlogos
blastn plot	Visualize the report of blastn	from gpro.evaluator.blast_plot import blastn_evaluation	blastn_evaluation

4.5 Utils

In addition, we provide some common functions in gpro.utils, such as reading fasta files, converting data formats, checking formats, disrupting and partitioning training sets. See Utils for detailed information.

Citations

[1] synthetic promoter design in e.coli based on deep generative models
[2] Argmax Flows and Multinomial Diffusion: Learning Categorical Distributions
[3] Deep flanking sequence engineering for efficient promoter design
[4] Vaishnav E D, de Boer C G, Molinet J, et al. The evolution, evolvability and engineering of gene regulatory DNA[J]. Nature, 2022, 603(7901): 455-463.
[5] Model-driven generation of artificial yeast promoters
[6] Zhang, P., Wang, H., Xu, H. et al. Deep flanking sequence engineering for efficient promoter design using DeepSEED. Nat Commun 14, 6309 (2023). https://doi.org/10.1038/s41467-023-41899-y
[7] Brookes D, Park H, Listgarten J. Conditioning by adaptive sampling for robust design\[C\]//International conference on machine learning. PMLR, 2019: 773-782.
[8] Gupta A, Zou J. Feedback GAN for DNA optimizes protein functions[J]. Nature Machine Intelligence, 2019, 1(2): 105-111.
[9] Vaishnav, E.D., de Boer, C.G., Molinet, J. et al. The evolution, evolvability and engineering of gene regulatory DNA. Nature 603, 455–463 (2022). https://doi.org/10.1038/s41586-022-04506-6
[10] Valeri J A, Soenksen L R, Collins K M, et al. BioAutoMATED: An end-to-end automated machine learning tool for explanation and design of biological sequences[J]. Cell Systems, 2023, 14(6): 525-542. e9.
[11] de Almeida B P, Reiter F, Pagani M, et al. DeepSTARR predicts enhancer activity from DNA sequence and enables the de novo design of synthetic enhancers[J]. Nature Genetics, 2022, 54(5): 613-624.
[12] de Almeida B P, Schaub C, Pagani M, et al. Targeted design of synthetic enhancers for selected tissues in the Drosophila embryo[J]. Nature, 2023: 1-2.