ALIEN

Research ALIEN

GitHub

Biologically informed AI

ALIEN Artificial Intelligence Explains geNetics

From genetic variation to biological insight.

ALIEN builds on GenNet to explore how biological knowledge, multiomics, and learning across cohorts can improve prediction and interpretation.

Explore the ALIEN framework How it works

Research vision · Built around GenNet

Explore the published foundations

Inside the framework

Conceptual example

Biology shapes the model.

A biologically structured neural network An illustrative network connects ten variants to five genes, three pathways and a phenotype prediction. Biological knowledge guides the links; interpretation examines contributions across the trained model. Biological knowledge Gene links · pathways · tissue context guides connectivity Variants Genes Pathways Prediction Phenotype Measured omics Separate input channels Interpret the trained model Gene and pathway contributions → hypotheses

Genetic variants feed gene and pathway layers. Biological knowledge defines the permitted connections.

Model connectionsBiological prior

How it works

A researcher would move through three choices. The diagram below is one illustration of the middle of that path.

  1. Choose data and biological knowledge

    Genotypes, measured omics, gene mappings, pathways, or reference atlases.

  2. Define and train a model

    Select a biological architecture and a training strategy.

  3. Investigate the output

    Assess prediction, explore gene and pathway contributions, and test hypotheses.

How knowledge changes the connections

Proximity, regulatory evidence, and tissue context are three different priors. Each lights a different set of links. This is an illustration, not an analysis.

Conceptual topology examples

Choose a component to read how it fits the model. Arrow keys move between components. These diagrams are illustrations, not an analysis tool.

Inputs

Variants

Genes

Pathways

Prediction

Model attribution can generate biological hypotheses. Replication and experimental validation are needed to assess them.

Published foundations and use cases

ALIEN brings together research on biologically structured neural networks, multiomics prediction, model interpretation and functional regulatory genomics. These publications include methodological foundations, a published GenNet application in Alzheimer’s disease genetics, and resources for a proposed population-to-clinical bridge.

Proposed ALIEN integration

Connecting population genomics and clinical genetics

ALIEN aims to connect insights from population cohorts with the interpretation of individual genetic variants through shared biological evidence. A proposed BRAIN-MAGNET integration would add brain-focused regulatory context to this workflow. The two research contexts share evidence while answering different questions.

Cohort scale

Population research

Cohort associations and gene or pathway signals.

Intended outputHypotheses for further study

Shared regulatory evidence

Regulatory annotations and candidate gene links, with their sources and score meanings preserved.

Individual variants

Clinical genetics research

Rare-variant candidates considered with phenotype, frequency and inheritance.

Intended outputCandidates for clinical review

Population research

Evidence used:

Intended output:

Clinical conclusions remain a separate review process. A population prediction or a model attribution is not a diagnosis.

BRAIN-MAGNET paper Explore BRAIN-MAGNET code Open this topic in the map

Built on GenNet. Designed to go further.

Available

GenNet foundation

GenNet is the open-source software for biologically structured neural networks. The repository is available now. ALIEN is a proposed workflow around that code — topology construction, experiments, benchmarking, and interpretation — and is not released as a separate framework.

  • Biological network topology
  • Phenotype prediction
  • Interpretation tools

Published foundations lists the studies behind this line. They do not mean the full ALIEN framework is released.

Explore GenNet

Proposed

Explore ALIEN

The map sets out proposed choices around GenNet: what could be configured, which questions would test those choices, and how a later step might recommend a configuration. That broader framework is not released.

  • Capabilities
  • Research questions
  • Future configuration
Explore ALIEN

Help shape biologically informed AI for genetics.

Explore GenNet on GitHub Gennady Roshchupkin’s website