International Business Machines Corporation

Matteo Manica

Maria Rodriguez Martinez

Seyed Ali Kazemi Oskooei

Raphael C Polig

Mitra Purandare

Roland Mathis

Jannis Born

Konstantinos Bekas

Antonio Foncubierta Rodriguez

Joris Cadow

A computer-implemented method for enforcing an idempotent-constrained characteristic during training of a neural network may be provided. The method comprises training of a neural network by minimizing a loss function, wherein the loss function comprises an additional term imposing an idempotence-based regularization to the neural network during the training.

Idempotence-constrained neural network

A machine learning model is generated for drug efficacy prediction in treatment of genetic disease from a dataset correlating gene expression data for disease-cell samples with drug efficacy values for the samples. Bias weights are stored that correspond to respective genes in the samples. Each bias weight is dependent on predetermined relevance of the respective gene to drug efficacy. The model is generated by processing the dataset via a tree ensemble method wherein decision trees are grown with splits corresponding to respective genes in the samples. The gene for each split is chosen from a respective subset of the genes, and genes are selected for inclusion in this subset with respective probabilities dependent on the corresponding bias weights. The model is stored, and can be applied to gene expression data measured for a patient to obtain a personalized drug efficacy prediction for devising a personalized course of treatment.

Drug efficacy prediction for treatment of genetic disease

Provide a reinforcement learning model including an agent and a critic; the critic includes a neural network pre-trained to generate, from input biomolecular data characterizing tissue cells and input compound data defining a compound molecule, a property value for said biomolecular action of that molecule on those tissue cells. The agent includes a neural network adapted to generate the compound data in dependence on input biomolecular data. Supply biomolecular data characterizing patient tissue cells to the agent and supply that data, and the compound data generated therefrom, to the critic to obtain a property value in an iterative training process in which reward values, dependent on the property values, are used to progressively train the agent to optimize the reward value. After training the agent, supply target biomolecular data, characterizing the target tissue cells, to the agent to generate compound data corresponding to a set of drug compounds.

Drug compound identification for target tissue cells

A neural network model is generated for drug efficacy prediction in treatment of genetic disease from a dataset correlating biomolecular data for disease-cell samples with drug efficacy values for a plurality of drugs, including defining training data pairs each including biomolecular data for a sample represented in the dataset and a string representation of a drug represented in the dataset; and using the training data pairs and the drug efficacy values corresponding to respective pairs to train weights of the model including a first attention-based encoder which encodes the biomolecular data of each pair, to produce encoded biomolecular data, a second attention-based encoder which encodes the string representation of each pair, to produce encoded string data, and a set of neural network layers which process the encoded biomolecular and string data for each pair, to produce a model output corresponding to drug efficacy.

Analysis of genetic disease progression may be provided. Data about a set of molecular status may be received. A dynamic prediction model of molecular interactions may be provided over time. The molecular statuses of the set over time may be determined using the dynamic prediction model. The determined molecular statuses may be clustered by applying an interaction-aware metric for the analysis of the genetic disease progression.

Interactive-aware clustering of stable states

An example of an embodiment is directed to a computer-implemented method for providing a cloud service to execute a computing task of a model specification. The method includes receiving, by the cloud service, the model specification and input data for the model specification from a user. The method further includes generating, by the cloud service, native code from the model specification and executing, by the cloud service, the computing task by executing the native code as a native process with the input data. The method also includes providing, by the cloud service, results of the computing task to the user. Other embodiments further concern a related computing system and a related computer program product.

Native code generation for cloud services

The present disclosure relates to a system and method for extracting information from text data. The method comprises: obtaining a plurality of text elements. A word embedding algorithm may be applied to the obtained text elements by mapping each text element of at least part of the text elements into a vector of a predefined dimension. The mapped text elements may be clustered into groups using the distances between the respective vectors. For each text element of a set of text elements of the mapped text elements a respective distribution of neighbors across the groups may be built. Similarity scores may be computed using the distributions thereby for determining relations between the set of text elements.

Growing community of inventors

Zurich, Switzerland

Matteo Manica