ChromBPNet models and data: Transcription factor stoichiometry, motif affinity and syntax regulate single cell chromatin dynamics during fibroblast reprogramming to pluripotency

<p>This record contains ChromBPNet models and data used to train the models for the paper &quot;Transcription factor stoichiometry, motif affinity and syntax regulate single cell chromatin dynamics during fibroblast reprogramming to pluripotency&quot; by Nair, Ameen <em>et al</em>.</p> <p>`data` contains bigwigs and regions (peaks + non-peaks) used for training each of the models. See `data/README.txt` for more details.</p> <p><strong>Models:</strong></p> <p><em>Loading the&nbsp;model:</em></p> <p>The models were trained using tf1.14. The models are provided in h5 format for tf1.14 (py3.7) and SavedModel format for tf2.X. tf2.X tested only for py3.8-11, tf2.8-13.</p> <p>To load the models in tf1.14:</p> <pre><code class="language-python">model = tf.keras.models.load_model("path/to/model.h5")</code></pre> <p>In tf2:</p> <pre><code class="language-python">model = tf.keras.models.load_model("path/to/model_dir")</code></pre> <p>If all fails, you can load the architecture as provided in `model_arch.py` with default parameters (`bpnet_seq` for bias model and `chrombpnet` for chrombpnet model), and then load the weights using `model.load_weights` from the weights provided in the `weights` directory.</p> <p>&nbsp;</p> <p><em>Usage:</em></p> <p>The bias models take as input one-hot sequence of length 2000. It has 2 outputs, a vector of logits of length 2000, and 1 logcounts scalar:</p> <pre><code class="language-python"># seq_one_hot of length B x 2000 x 4 out_bias_logits, out_bias_logcounts = bias_model.predict(seq_one_hot) # out_bias_logits: B x 2000 # out_bias_logcounts: B x 1</code></pre> <p>The ChromBPNet model takes as input a one-hot sequence of length 2000, bias logits of length 2000 and bias log-counts scalar. It has the same output types as the bias model. To run the chrombpnet model to obtain predictions:</p> <pre><code class="language-python">pred_profile, pred_logcounts = chrombpnet_model.predict([seq_one_hot, out_bias_logits, out_bias_logcounts]) # pred_profile: B x 2000 # pred_logcounts: B x 1 </code></pre> <p>If you wish to obtain the &quot;de-biased&quot; predictions (see Methods), simply pass in zeros instead of the bias model predictions as:</p> <pre><code class="language-python">pred_profile_debiased, pred_logcounts_debiased = chrombpnet_model.predict([seq_one_hot, np.zeros((seq_one_hot.shape[0], 2000)), np.zeros((seq_one_hot.shape[0], 1))])</code></pre> <p>To obtain predicted per-base predicted counts (with or without bias):</p> <pre><code class="language-python">pred_per_base_counts = scipy.special.softmax(pred_profile, axis=-1) * (np.exp(pred_logcounts)-1) # pred_per_base_counts: B x 2000 </code></pre> <p>Note that in general predicted counts can&#39;t be compared across models as they are not corrected for sequencing depth.</p> <p>&nbsp;</p> <p><em>Note:</em></p> <p>All bias models used across folds are identical, except for the final intercept term in the counts output (see Methods), that is specific to each cell state, fold combination.</p> <p>&nbsp;</p> <p><em>Folds:</em></p> <p>The splits used for training the different folds are as below:</p> Fold Test Chromosomes Validation Chromosomes 0 chr1 chr8, chr10 1 chr2, chr19 chr1 2 chr3, chr20 chr2, chr19 3 chr6, chr13, chr22 chr3, chr20 4 chr5, chr16, chrY chr6, chr13, chr22 5 chr4, chr15, chr21 chr5, chr16, chrY 6 chr7, chr18, chr14 chr4, chr15, chr21 7 chr11, chr17, chrX chr7, chr18, chr14 8 chr9, chr12 chr11, chr17, chrX 9 chr8, chr10 chr9, chr12 <p>Remaining chromosomes were used as the training chromosome for each fold.</p>