Differences

This shows you the differences between two versions of the page.

--- pathway_prediction_in_batch_mode [2025/04/28 13:10] – shankar
+++ pathway_prediction_in_batch_mode [2025/05/22 09:53] (current) – shankar
@@ Line 8: / Line 8: @@
-  * Input file: A .tsv file containing SMILES strings and compound names,** without a header**. A template input file can be downloaded [[https://drive.switch.ch/index.php/s/ZGxStHlMTITB7ys|here]].
+  * **Input file**: A .tsv file containing SMILES strings and compound names,** without a header**. A **template input file** can be downloaded [[https://drive.switch.ch/index.php/s/ZGxStHlMTITB7ys|here]]. There is an upper limit of **250 compounds per file**. If your input file contains more than 250 compounds, please split it into smaller batches before running the predictions.
-  * Relative Reasoning model: The relative reasoning model that will be used for the predictions. A default model is pre-selected, but users can     also choose a different model if preferred.
+  * **Relative Reasoning model**: The relative reasoning model that will be used for the predictions. A default model is pre-selected, but users can     also choose a different model if preferred.
-  * Number of transformation products (TPs): The maximum number of TPs to predict per input compound.
+  * **Number of transformation products (TPs)**: The maximum number of TPs to predict per input compound. The default value is set to 30 TPs per compound, but users can choose any number between 1 and 50.
+==== Algorithm Description ====
+The batch mode uses a greedy-search algorithm where:
+  * Compounds are represented as Nodes
+  * Biotransformation reactions are represented as edges, each assigned a weight corresponding to the predicted probability of the reaction occurring, based on available data and competing reaction pathways.
+The probability of a reaction (p_edge) is obtained from the machine learning-based relative reasoning algorithm.
+The probability of a child node generated during pathway search, also known as the **combined probability**, is determined based on the probability of the parent node and the reaction probability, as illustrated in the below figure.
+{{ :batch_mode_edited.png |}}
+All transformation products (TPs) resulting from reactions with a probability greater than zero are stored in a priority queue, ordered in descending order based on their combined probability.
+The greedy algorithm makes a locally optimal decision at each step by always expanding the node with the highest combined probability first.
+Pathway exploration continues until either the user-defined threshold for the number of TPs is reached, or there are no more TPs remaining in the priority queue to expand.
+In some cases, compounds may have fewer predicted TPs than the user-defined threshold, or no TPs predicted at all.
+This can happen for either of the following reasons:
+  * No applicable transformation rules were available beyond a certain point, causing the pathway search to stop.
+  * The probabilities for all predicted transformation products were zero, preventing further expansion.