Services

The following are some specific AI services under development in the bio-data track. Some will be released with the beta version of Osiris in February 2019 and some are slated for later release based on ongoing work.

Supervised Classification of Binary-Valued Data Using MOSES

The Osiris agent accepts a data file with/and classification label information and program algorithm and validation parameters and returns a file of scored combo models and a ranked list of model features. User Interface for Supervised Classification of SNP or Gene-Expression Data Using MOSES A web-based interface accepts a data file input with category labels, provides an interface for setting algorithm and validation parameters, optionally solicits an BSC wallet address, and returns a file containing scored combo models and a ranked list of model features generated by the MOSES.

Annotation of MOSES Results, or other Gene Sets, Using AtomSpace Knowledge Base

A web-based interface accepts a list of gene names or reference IDs; optionally provides a screen to select from a list of reference knowledge bases, annotation types, and filtering parameters; and returns a table of the input genes and their annotations and/or a graph representation of the input genes and their annotations in a selected standard graph format.

Symbolic Regression on Genetics Datasets

The Osiris agent accepts a genetic data package consisting of a genetic and numerical biomarker dataset, numerical outcome values associated with each sample, and program algorithm and validation parameters. It outputs a results file containing a model that predicts the phenotype number corresponding to that genetic data package. Optionally, either FFX or MOSES algorithms can be indicated by the user.

Textual User Interface for Querying Result Sets or Knowledge Bases

A natural language query parser based on GHOST will allow context-dependent queries, given an AtomSpace, with selected knowledge bases and analysis results as input.

Bio-NLP Textual Relationship Extraction

Using existing open-source tools to tag bioentities (small molecules, genes, proteins, cell types, organisms, diseases.), OpenCog natural language processing tools will extract relations among them from arbitrary plain text or pdf documents and output an AtomSpace representation of these relationships. AtomSpace knowledge bases will be updated with new information. These knowledge bases are useful for data mining and inference processes related to user investigations.

Transfer Learning from Model Organism Knowledge Bases

One of the major challenges in genetics is to predict the functions of genes and proteins and to identify their regulatory pathways. Data mining and several machine learning techniques have been successfully applied to transfer gene annotation information between organisms.

Cell-level Hypothesis Generation from ML Results Given AtomSpace Knowledge Base and Genome-scale Cell-Metabolism Model

Given a feature list of variants, transcript expression levels, and/or protein abundances; a cell type and other context from experimental results data; and an AtomSpace containing background knowledge from public or proprietary customer sources, causal hypotheses are generated to explain the observed phenotypes associated with experimental data.

Tissue-level Hypothesis Generation from ML Results Given AtomSpace Knowledge Base and Cell Ensemble Model Including Extracellular Environment

Knowledge-base contents and inference rule bases are combined with extracellular and tissue-level context to allow us to generate meaningful hypothesis-driven inferences based on clinical and laboratory parameters of experimental sample subjects.