Predmoter models

The listed models are trained Predmoter (https://github.com/weberlab-hhu/Predmoter) models. Out of three replicates, meaning three training runs with different seeds, these models had the highest validation "accuracy", meaning Pearson's correlation. The models were trained on plant ATAC- and/or ChIP-seq (H3K4me3) data.

IMPORTANT: Models trained on a GPU can be used to generate predictions on the CPU. The CPU predictions will be the same as the GPU predictions would be. Predicting on the CPU will take a little longer.

The best model for predicting ATAC- and H3K4me3 ChIP-seq seq data is the Combined_02 model.

Table of contents

1. Model information
2. Model performance
3. Predict with Predmoter
4. Additional information
5. References
6. Citation

1. Model information

A model can only predict the dataset(s) it was trained on. The model architecture is listed for completeness.

Model name	Dataset(s)	Architecture
U-Net	ATAC-seq	3 convolutional layers + 3 transposed convolutional layers
Hybrid	ATAC-seq	U-Net + 2 LSTM layers
BiHybrid	ATAC-seq	U-Net + 2 BiLSTM layers
BiHybrid_02	ATAC-seq	U-Net + 2 BiLSTM layers + 6 batch normalization layers
BiHybrid_03.1	ATAC-seq	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.3)
BiHybrid_03.2	ATAC-seq	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.5)
BiHybrid_04	ATAC-seq, filtered flagged sequences*	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.3)
BiHybrid_05	ChIP-seq (H3K4me3), filtered flagged sequences*	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.3)
Combined	ATAC-seq, ChIP-seq (H3K4me3), filtered flagged sequences*	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.3)
Combined_02	ATAC-seq, ChIP-seq (H3K4me3), filtered flagged sequences*, additional training data	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.3)
IS_10	ATAC-seq, ChIP-seq (H3K4me3), filtered flagged sequences*, additional training data; intra-species training and validation split (validation set: 10%)	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.3)
IS_20	ATAC-seq, ChIP-seq (H3K4me3), filtered flagged sequences*, additional training data; intra-species training and validation split (validation set: 20%)	U-Net + 2 BiLSTM layers + 6 batch normalization layers + 1 dropout layer (dropout probability of 0.3)

*excluded subsequences of unplaced scaffolds and non-nuclear sequences during training and testing

2. Model performance

The performance of all models is shown for the two test species Arabidopsis thaliana and Oryza sativa. The figures can also be found in the paper (figure 5, 6, and 7).

Performance of the best models per model setup and test species +/- 3 kbp around all TSS: (a) The average predicted read coverage for each model and dataset of both strands was compared to the average experimental read coverage of both strands via Pearson’s correlation. (b) The average experimental read coverage (target/Y) and predicted ATAC- and ChIP-seq read coverage in reads per bp are shown for A.thaliana and O.sativa. The predictions of five of the nine best ATAC-seq models and of all three best ChIP-seq models are depicted. Flagged sequences were excluded from the calculations (see Section 2.1.2 of the paper).

Peak F1 statistics and precision-recall curves: (a) The F1 of the predicted peaks versus the experimental peaks was calculated per model, test species and NGS dataset. (b) The precision-recall curves were calculated per test species, model, and dataset. The threshold/cutoff points are marked by circles. The exact parameters and cutoffs used are listed in Section 2.4 of the paper. Two AUPRC baselines (dashed lines) are depicted. Baseline 2 only applied to the Combined_02 model’s predictions, as this model trained on additional data (see Section 2.3.1 of the paper). (c) The Combined_02 model’s predictions were smoothed utilizing a rolling mean with a given window size (ws). The precision-recall curves were calculated per test species, window size, and dataset. The threshold/cutoff points are marked by circles. The exact parameters and cutoffs used are listed in Section 2.4 of the paper. The AUPRC baseline (dashed line) is depicted. Flagged sequences were excluded from peak calling and F1, precision and recall calculations (see Section 2.1.2 of the paper).

Example predictions of Predmoter: Example regions comparing Predmoter’s unsmoothed predictions to experimental data (target/Y) for the test species (a, b) A. thaliana and (c, d) O. sativa in 5’ to 3’ direction are depicted. The plots (a) and (c) show the ATAC-seq read coverage per bp, the plots (b) and (d) show ChIP-seq (H3K4me3) read coverage per bp.

3. Predict with Predmoter

Example of how to predict with Predmoter on a fasta file using one of the models above (The batch size depends on the capacity of the CPU/GPU. The default batch size is 120. If the error RuntimeError:CUDA out of memory. or other memory errors occur, try setting a lower batch size.):

Predmoter.py -f <species>.fasta -o <output_directory> -m predict \
--model <path_to_model>/Combined_02_predmoter_v0.3.2.ckpt -b <batch_size> \
-of bigwig --species <species>

For more information, please refer to this subsection of the Predmoter documentation.

4. Additional information

For additional information about training, validation and test data, species selection and exact hyperparameter configuration used, please refer to the supplementary material of the paper (see citation).

References

Zhang, Y., Liu, T., Meyer, C. A., Eeckhoute, J., Johnson, D. S., Bernstein, B. E., Nussbaum, C., Myers, R. M., Brown, M., Li, W., Shirley, X. S. (2008). Model-Based Analysis of ChIP-Seq (MACS). Genome Biology, 9(9) , 1–9. https://doi.org/10.1186/GB-2008-9-9-R137

Citation

Kindel, F., Triesch, S., Schlüter, U., Randarevitch, L.A., Reichel-Deland, V., Weber, A.P.M., Denton, A.K. (2024) Predmoter—cross-species prediction of plant promoter and enhancer regions. Bioinformatics Advances, 4(1), vbae074. https://doi.org/10.1093/bioadv/vbae074