Scientists at the Wyss Institute for Biologically Inspired Engineering at Harvard University and MIT have created an automated machine learning (AutoML) platform called BioAutoMATED, designed specifically for biologists with little to no machine learning experience. The platform allows easy analysis, design, and interpretation of biological sequences, such as nucleic acids, peptides, and glycans, enabling researchers to uncover valuable insights without extensive coding. The study detailing the platform's capabilities was published in Cell Systems.
The field of biology has seen an exponential increase in the amount of data generated, thanks to advancements in sequencing technology and computing power. However, extracting meaningful information from this vast amount of data has proven challenging. Machine learning and artificial intelligence tools can greatly accelerate the data analysis process, but the complexity and technical requirements of most ML tools often hinder non-experts from utilizing their potential.
BioAutoMATED aims to bridge this gap by providing biologists with a comprehensive AutoML platform that requires minimal user input and no prior ML experience. Co-first author Jackie Valeri, a graduate student at the Wyss Institute, emphasizes the tool's purpose: to empower biologists and researchers from various domains to leverage ML and AutoML techniques to address fundamental questions and uncover significant biological insights.
The inspiration for BioAutoMATED arose from a shared realization among the team members that the accessibility of ML tools for biologists was limited. By developing this user-friendly platform, the researchers aimed to drive the AI biotech revolution within the biological research community, leveraging the expertise of biologists themselves.
While existing AutoML systems exist to simplify the generation of ML models from datasets, they often have limitations, such as focusing on specific model types and lacking automation features for biological sequences. To overcome these challenges, the team modified three existing AutoML tools: AutoKeras, DeepSwarm, and TPOT. By integrating these approaches, BioAutoMATED generates standardized output results that allow users to compare and select the most useful insights for their data.
BioAutoMATED accepts input data in the form of DNA, RNA, amino acid, and glycan sequences of any length or biological function. It automatically preprocesses the input data and generates models that predict biological functions solely based on sequence information. The platform also includes features that help users identify the need for additional data, determine the most relevant sequence features, and design new sequences for future experiments.
To demonstrate the capabilities of BioAutoMATED, the researchers conducted several experiments. In one study, they investigated the impact of changing the sequence of an RNA stretch known as the ribosome binding site (RBS) on translation efficiency in E. coli. The platform accurately predicted translation efficiency using the DeepSwarm algorithm, generating results comparable to those produced by ML experts but with significantly reduced coding requirements.
The team further explored the application of BioAutoMATED in analyzing peptide and glycan sequences. They successfully identified key amino acids in a peptide sequence that influence an antibody's ability to bind to the drug ranibizumab (Lucentis) and classified different types of glycans based on immunogenicity.
Although any models generated using BioAutoMATED require experimental validation, the researchers envision its integration into the expanding suite of AutoML tools. They anticipate that the platform's functionality could extend beyond biological sequences to other sequence-like objects, such as fingerprints.
By democratizing machine learning and making it accessible to non-experts, BioAutoMATED has the potential to revolutionize the way biologists uncover the intricacies of life and accelerate scientific discoveries.
The field of biology has seen an exponential increase in the amount of data generated, thanks to advancements in sequencing technology and computing power. However, extracting meaningful information from this vast amount of data has proven challenging. Machine learning and artificial intelligence tools can greatly accelerate the data analysis process, but the complexity and technical requirements of most ML tools often hinder non-experts from utilizing their potential.
BioAutoMATED aims to bridge this gap by providing biologists with a comprehensive AutoML platform that requires minimal user input and no prior ML experience. Co-first author Jackie Valeri, a graduate student at the Wyss Institute, emphasizes the tool's purpose: to empower biologists and researchers from various domains to leverage ML and AutoML techniques to address fundamental questions and uncover significant biological insights.
The inspiration for BioAutoMATED arose from a shared realization among the team members that the accessibility of ML tools for biologists was limited. By developing this user-friendly platform, the researchers aimed to drive the AI biotech revolution within the biological research community, leveraging the expertise of biologists themselves.
While existing AutoML systems exist to simplify the generation of ML models from datasets, they often have limitations, such as focusing on specific model types and lacking automation features for biological sequences. To overcome these challenges, the team modified three existing AutoML tools: AutoKeras, DeepSwarm, and TPOT. By integrating these approaches, BioAutoMATED generates standardized output results that allow users to compare and select the most useful insights for their data.
BioAutoMATED accepts input data in the form of DNA, RNA, amino acid, and glycan sequences of any length or biological function. It automatically preprocesses the input data and generates models that predict biological functions solely based on sequence information. The platform also includes features that help users identify the need for additional data, determine the most relevant sequence features, and design new sequences for future experiments.
To demonstrate the capabilities of BioAutoMATED, the researchers conducted several experiments. In one study, they investigated the impact of changing the sequence of an RNA stretch known as the ribosome binding site (RBS) on translation efficiency in E. coli. The platform accurately predicted translation efficiency using the DeepSwarm algorithm, generating results comparable to those produced by ML experts but with significantly reduced coding requirements.
The team further explored the application of BioAutoMATED in analyzing peptide and glycan sequences. They successfully identified key amino acids in a peptide sequence that influence an antibody's ability to bind to the drug ranibizumab (Lucentis) and classified different types of glycans based on immunogenicity.
Although any models generated using BioAutoMATED require experimental validation, the researchers envision its integration into the expanding suite of AutoML tools. They anticipate that the platform's functionality could extend beyond biological sequences to other sequence-like objects, such as fingerprints.
By democratizing machine learning and making it accessible to non-experts, BioAutoMATED has the potential to revolutionize the way biologists uncover the intricacies of life and accelerate scientific discoveries.