INFINITE SERIES: AI driven computational systems

  • AI driven drug development is becoming the future. The future is already here. We will provide customers the online computing system, including our DLEPS, to facilitate the drug discovery pipelines.
  • Users can explore the chemical space in an interesting deep learning perspective. Further, users can design new molecules under the guide of our AI systems.
  • Synthetic accessibility, toxicity, activity, efficacy, long-term adversarial effect were estimated by AI and will guide users for drug design.
  • PCC License Out

  • We already opened pipelines and obtained exciting results in obesity, NASH, hyperuricemia, longevity, coronavirus, osteoporosis and etc.
  • We're good at the traditionally defined complex diseases. Contact us if you’re interested.
  • Introduction of DLEPS

    For the CTPs' sub-system, the current version is able to predict the CTPs with mean correlation coefficient is 0.74, while the peak is at 0.9. 68% of molecules have r>0.74. Using DLEPS, we were able to find candidates for various diseases, including obesity, hyperuricemia, aging, NASH and etc.

  • DLEPS can be iteratively applied for progressively improve the efficacy.
  • DLEPS does not depend on know targets of diseases.
  • DLEPS can be used to find novel targets for diseases by find the compound first, than study their MOA.
  • Online DLEPS system

    DLEPS: Deep Learning based Efficacy Prediction System (DLEPS)

    Target based drug development has made big progress in various diseases. Yet, targets for a lot of diseases have not been identified. NLP based literature mining depend on previous studies was limited. Previous data have not been designed for deep learning training thus they're not fittable. Well organized data should be generated.

    Through comprehensively exploration of biological data and omics data of diseases, we developed DLEPS to unravel such hidden links.

    Deep neural networks have infinite power to fitting massive data set of small molecules thus extrapolate the measured links to enormous virtual molecules.

    InfiniteBind: Ligand Ensembles based small molecules’ design

    A set of small molecules of a particular target resembles the spatial confirmation of the target and thus can be used to predict new molecules through various machine learning classifiers. Small molecules can be encoded in a high dimensional space by variational autoencoders through deep neural network learning.

    Monte Carlo sampling in such space yield transitional change of molecules. Simulated Annealing guided by specific classifiers will lead us iteratively design of molecules with improved binding probability. We have applied such technique to NaV1.7 for a proof-of-concept demonstration.

    InfiniteDock: an intelligent docking system driven by Reinforcement Learning

    Brutal force docking screening is one of the current methods for hit discovery. The most efficient system was able to sampling about 10 million small molecules for one target. But, such a number is a very small subset of total virtual molecules. Reinforcement Learning provide us an efficient guided sample route towards better hit in the encoding space.

    MetaRetro: a retrosynthesis prediction system for natural products

    Search an enzymatic pathway for synthesis of natural products. Deep learning can help to find the most plausible pathway for synthesis of particular products. Our Hyper Cube Shrink Algorithm revealed the causal relationship between enzymatic activity and flux re-distribution. Taken together, both design strategies and optimization rules can be found by combing those two platforms.