DeepProphet: The state-of-the-art machine learning techniques serving Life Science Research
GeneRecommender is based on a proprietary neural network, called DeepProphet, scanning millions of scientific articles and processing about 445 million connections. This engine is able, given a set of genes and/or diseases in which the user is interested, to suggest a new set of genes that are related and should be considered by the users. We integrated data from several public DBs to exploit the most updated information.
AN UNSTOPPABLE GROWING PLATFORM BASED ON VALUABLE DATA
Millions of scientific papers have been used to train GeneRecommender's Artificial Intelligence engine. During the learning process, the Artificial Neural Network considered what life science researchers worldwide have produced in the past 30 years to extract patterns and correlations among genes and diseases.
Grow your knowledge THANKS TO ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
The Algorithm - DeepProphet2
GeneRecommender uses DeepProphet2, an advanced deep learning algorithm trained on all the scientific papers available online. Indeed, using cutting-edge Natural Language Processing (NLP) techniques, the system extracts key information from manuscripts and uses it to train a deep learning neural network to predict which genes might be associated with a given study. In order to accomplish this task, TheProphetAI developed a customer transformer-based model based on a design similar to that of the most famous and successful language models, such as LaMDA and GPT-3. We have tested the System using gene sets representing pathways and diseases. Recursively, a randomly chosen gene was left out and the remaining input was provided, we tested whether the algorithm could predict which gene was missing based on its first recommendation. Since the algorithm does not know the gene sets during training, if it can complete them, it has learned an accurate representation of genes. Using this kind of analysis, we reached an astonishing result: an AUC of 0.9.
If you would like to go into technical details about the algorithm and its validation, please consider downloading our scientific paper: [https://arxiv.org/abs/2208.01918]
Human and AI cooperation
You can greatly benefit from the introduction of Artificial Intelligence systems into your workflow. In order to maximize these technologies’ potential, AI must be adopted safely and appropriately. While AI cannot replace humans in their jobs, it can help and support them in their daily tasks, improving their efficiency and effectiveness. The predictions made by AI systems aren’t always right, but they offer a useful starting point for solving specific issues.
Our belief is that the real potential lies in the cooperation between humans and machines. Deep analysis by experts is required to understand the real value of the recommendations obtained from the AI engine.
While you can find tools on our platform to understand the results of each recommendation, please keep in mind that we humans use AI because we cannot work with such data!
Do the results always turn out correctly? Although it may not be possible every time, if you insert a sufficient number of targets, you are likely to see some valuable results statistically. Most of the recommended genes will provide you with some useful insights for your research.
The process that Artificial Intelligence follows is the same as the process Humans follow in a fraction of the time. Researchers like you often use scientific literature to learn about what has been done before on similar topics and to better understand their own research. Indeed, we are using exactly the same data, elaborated by a powerful neural network that knows what the scientific community has already studied and tested and can use this knowledge to generate recommendations.
AI will not substitute your work, but it will help you to speed up your studies with its innovative and crucial insights.