THE FIRST BIOMEDICAL SPECIFIC MACHINE LEARNING FRAMEWORK

Augusta is a biomedical AI (Artificial Intelligence) and ML (Machine Learning) framework designed to transition time from data pre-processing and integration to model building and interrogation using familiar toolsets within Python. Augusta begins with diverse, raw medical data types (e.g. images, chemical structures, genomic data, tabular data), and operates across three modules:

  1. Augusta Pre-Processing
  2. Augusta ML (Machine Learning)
  3. Augusta Architect
Common Use Cases:
  • Drug discovery and development, incl. small molecule activity prediction
  • Diagnostics & precision medicine
  • Patient outcomes prediction and stratification
Capabilities:
  • Preprocessing
  • Feature reduction & selection
  • Data Integration (e.g. combining genomics with clinical data)
  • Model creation
  • Model tuning
  • Model training
  • Model interrogation
  • Visualization
Benefits:
  • Faster, effective data pre-processing, directly integrated with model building
  • Seamless distributed computing
  • Flexible architecting of processing pipelines, changing as data type and volume requires
  • Adaptable to changing needs/preferences over time

AUGUSTA PRE-PROCESSING

  • Quickly standardize or normalize data
  • Permute over pre-processing options
  • Create workflows where critical biases can be reduced
  • Save, modify, and re-run workflows

Use with:

  • Any Data Source: BYOD (Bring Your Own Data) local, databases, cloud
  • Any Combination of Sources: Modular and customizable pipelines for processing raw data in any combination

Data Integration (sample pipelines)

  • MRI/fMRI and other imaging modalities
  • EEG
  • EKG/ECG
  • Genomics,
  • Proteomics
  • Chemistry
  • EHR/EMR data
  • Streaming/wearables data
  • Tabular data
  • Custom data options available

Feature Optimization

Integrating data of various types (e.g. combining genomics with clinical data), enables the engineering of unique features, providing for greater machine learning insights. Features can be easily grouped, sub grouped, and archived, making them easily accessible to models, increase tuning parameters, and enhanced interrogation capabilities

AUGUSTA ML

Model Creation

Use machine learning models from Tensorflow and Scipy, with a unified syntax and output

Model Tuning and Interrogation

Iterate over model-specific parameters, investigate the impact of combinations of pre-processing decisions and model hyperparameters in the context of model performance

  • Evaluate model performance via cross-validation, using metrics such as accuracy, precision/recall and AUC
  • Implement multiple feature reduction methods and evaluate impact on model performance
  • Visualization incorporating Seaborn and Matplotlib packages

Contingent-AI™ (Patent pending) allows data scientists to permute options in the model generation process based on decisions made in pre-processing.

AUGUSTA ARCHITECT

Augusta Architect is a simple, Python-based syntax that allows the processing and integration of multiple, diverse data types, and ability to run/compare multiple machine learning algorithms

Augusta Architect uses “code blocks” to construct the data flows and schematic framework for preprocessing and machine learning, reducing the time required to program the flow from data collection through your machine learning engine.  A single toolset from data ingestion to result output, eliminates the need to port data from system to system.  Augusta effectively maintains data integrity and eliminates error prone steps.  The result:

  • Increased speed to market
  • Easy iteration and edits
  • Stronger confidence
  • Greater precision
  • Lower cost of R&D efforts
Request a Demo
Augusta™ Architect is LIVE!
Augusta™ Video
Watch time  2:min
Request a Demo
Case Studies
June 16, 2020 in Case Study, Drug Discovery, Media, Video

Predicting Mechanism of Action Using AI – Webinar

In this webinar, Mikalai Malinouski Ph.D.demonstrated how BioSymetrics uses AI in the prediction of mechanism of action (MOA), and provided a an example of how machine learning workflows can benefit…
Read More
May 11, 2020 in Case Study, Drug Discovery

Poster Presentation: MIT 2020 AI for Drug Discovery Conference

Some of our high level learning was recently presented at the 2020 MIT AI for Drug Discovery conference in February. (more…)
Read More
May 4, 2020 in Article, Blog Post, Case Study, Media, Uncategorized

MOA Prediction Platform to Help With COVID-19 Research

BioSymetrics Mechanism of Action Prediction Platform framework will be put to work both in prioritizing chemical libraries and in predicting mechanism for identified compounds. Per the three COVID-19 research initiatives…
Read More
March 23, 2020 in Blog Post, Case Study, Drug Discovery

Mitigating Batch Effects in Cell Painting Data

With the advent of high content screening methodologies (e.g. cellular imaging, transcriptomics, etc.), it becomes more challenging to tease apart and visualize batch effects. This is further compounded when building…
Read More
Request a Demo
Webinar (Archive)
Click image to view. Watch time 20:min
White Paper