Interaction network inference from vector representation of words

A computer program product comprising a computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code configuring a computer system to:

• obtain from a plurality of documents, a plurality of text elements comprising names of molecular entities;
• apply a word embedding algorithm to the obtained text elements comprising mapping each text element of at least part of the text elements into a vector of a predefined dimension, wherein to map each text element, said computer-readable program code further configuring said computer system to:
• receive a text element input;
• train a model to predict the surrounding words of the input text element, said predict model training comprising:
• define a window of predetermined size spanning an amount of words surrounding the text element to obtain a word context from a document,
• map a target word to its surrounding words within the window;
• feed pairs of target-neighboring words into a neural network, said neural network trained to optimize a probability of a semantic meaning of a word given the obtained word context;
• and
• represent said received text element according to a real vector to capture said semantic meaning;
• cluster the mapped text elements into groups using the distances between the respective vectors;
• build for a set of text elements of the mapped text elements a respective distribution of neighbors across the groups;
• and
• compute similarity scores using the distributions thereby determining relations between the set of text elements;
• build a molecular interaction network using the similarity scores, wherein to build the molecular interaction network, said computer-readable program code further configuring said computer system to:
• provide nodes corresponding to one or more text elements of the set of text elements in a graph structure;
• connect edges between the nodes;
• and
• weigh the edges by the similarity score values of the nodes they connect, an edge weight representing a confidence of each interaction between the molecules linked by the edge.

A computer system to extract information from text data, the computer system comprising:

• a processor;
• and
• a non-transitory computer-readable memory coupled to the processor, wherein the memory comprises instructions which, when executed by the processor, cause the processor to:
• obtain from a plurality of documents, a plurality of text elements comprising names of molecular entities;
• apply a word embedding algorithm to the obtained text elements comprising mapping each text element of at least part of the text elements into a vector of a predefined dimension, wherein to map each text element, said hardware processor is further configured to:
• receive a text element input;
• train a model to predict the surrounding words of the input text element, said predict model training comprising:
• define a window of predetermined size spanning an amount of words surrounding the text element to obtain a word context from a document,
• map a target word to its surrounding words within the window;
• feed pairs of target-neighboring words into a neural network, said neural network trained to optimize a probability of a semantic meaning of a word given the obtained word context;
• and
• represent said received text element according to a real vector to capture said semantic meaning;
• cluster the mapped text elements into groups using the distances between the respective vectors;
• build for a set of text elements of the mapped text elements a respective distribution of neighbors across the groups;
• compute similarity scores using the distributions thereby determining relations between the set of text elements, and
• build a molecular interaction network using the similarity scores, wherein to build said interaction network between molecular entities of interest, the processor is further configured to perform:
• providing nodes corresponding to one or more text elements of the set of text elements in a graph structure;
• connecting edges between the nodes;
• and
• weighing the edges by the similarity score values of the nodes they connect, an edge weight representing a confidence of each interaction between the molecules linked by the edge.