A drug library management system generates versions of drug library data that can be used by infusion pumps, and version of drug library data that can be used by systems or components in a clinical environment other than infusion pumps. One version of the drug library data may be customized for a particular infusion pump, while another version may be a generalized version that can be used by middleware systems that process messages received from various infusion pumps that are using a different version of the drug library data. The generalized version may be archived separately from a drug library database used by the drug library management system to generate the various versions.

A drug library management system generates versions of drug library data that can be used by infusion pumps, and version of drug library data that can be used by systems or components in a clinical environment other than infusion pumps. One version of the drug library data may be customized for a particular infusion pump, while another version may be a generalized version that can be used by middleware systems that process messages received from various infusion pumps that are using a different version of the drug library data. The generalized version may be archived separately from a drug library database used by the drug library management system to generate the various versions.

A variational autoencoder (VAE) has been developed to learn a continuous numerical, or latent, representation of molecular structure to expand reference libraries for small molecule identification. The VAE has been extended to include a chemical property decoder, trained as a multitask network, to shape the latent representation such that it assembles according to desired chemical properties. The approach is unique in its application to metabolomics and small molecule identification, focused on properties that are obtained from experimental measurements (m/z, CCS) paired with its training paradigm, which involves a cascade of transfer learning iterations. First, molecular representation is learned from a large dataset of structures with m/z labels. Next, in silico property values are used to continue training. Finally, the network is further refined by being trained with the experimental data. The trained network is used to predict chemical properties directly from structure and generate candidate structures with desired chemical properties. The network is extensible to other training data and molecular representations, and for use with other analytical platforms, for both chemical property and feature prediction as well as molecular structure generation.

The subject matter described herein includes computational methods, systems and non-transitory computer readable media for de-novo drug discovery, which is based on deep learning and reinforcement learning techniques. The subject matter described herein allows generating chemical compounds with desired properties. Two deep neural networksgenerative and predictive, represent the general workflow. The process of training consists of two stages. During the first stage, both models are trained separately with supervised learning algorithms, and during the second stage, models are trained jointly with reinforcement learning approach. In this study, we conduct a computational experiment, which demonstrates the efficiency of proposed strategy to maximize, minimize or impose a desired range to a property. We also thoroughly evaluate our models with quantitative approaches and provide visualization and interpretation of internal representation vectors for both predictive and generative models.

The subject matter described herein includes computational methods, systems and non-transitory computer readable media for de-novo drug discovery, which is based on deep learning and reinforcement learning techniques. The subject matter described herein allows generating chemical compounds with desired properties. Two deep neural networksgenerative and predictive, represent the general workflow. The process of training consists of two stages. During the first stage, both models are trained separately with supervised learning algorithms, and during the second stage, models are trained jointly with reinforcement learning approach. In this study, we conduct a computational experiment, which demonstrates the efficiency of proposed strategy to maximize, minimize or impose a desired range to a property. We also thoroughly evaluate our models with quantitative approaches and provide visualization and interpretation of internal representation vectors for both predictive and generative models.

The disclosure provides methods and systems for identifying a subset of compounds in a plurality of compounds. The identifying includes obtaining, for each compound, a vector including a set of elements, where each element includes a measurement of a different feature of an instance of a cell context upon exposure to the compound. The identifying includes repeating the obtaining for a plurality of cell contexts, to obtain a plurality of vectors for each compound across different cell contexts. The identifying includes combining the vectors for each compound to form a combined vector for each compound, thereby forming a plurality of combined vectors representing different compounds. The identifying includes pruning the plurality of compounds to the subset of compounds based on a similarity between respective combined vectors in the plurality of combined vectors corresponding to compounds in the plurality of compounds.

A method of designing a D-polypeptide that binds with an L-target protein can include: identifying a polypeptide target having L-chirality; determining hotspot amino acids of a polypeptide ligand having L-chirality that have binding interactions with the L-target protein; determining transformations of side chains of the hotspot amino acids that retain the binding interactions with the target; generating inversed hotspot amino acids with chirality opposite to the one of the target; identifying a polypeptide having inverse chirality from the target protein, on which a combination of inversed hotspot amino-acid can be grafted without significantly changing their interactions with the target. The designed ligands can be processed and converted to D-ligands that bind with the L-target protein.

A method of designing a D-polypeptide that binds with an L-target protein can include: identifying a polypeptide target having L-chirality; determining hotspot amino acids of a polypeptide ligand having L-chirality that have binding interactions with the L-target protein; determining transformations of side chains of the hotspot amino acids that retain the binding interactions with the target; generating inversed hotspot amino acids with chirality opposite to the one of the target; identifying a polypeptide having inverse chirality from the target protein, on which a combination of inversed hotspot amino-acid can be grafted without significantly changing their interactions with the target. The designed ligands can be processed and converted to D-ligands that bind with the L-target protein.

A composition-of-matter comprising a crystallized form of a large ribosomal (50S) subunit of a pathogenic bacterium, and the atomic coordinates of the three-dimensional structure thereof are provided herein, as well as methods for crystallizing the same, and using the atomic coordinates of the same to design de novo ligands with high specificity thereto.

A composition-of-matter comprising a crystallized form of a large ribosomal (50S) subunit of a pathogenic bacterium, and the atomic coordinates of the three-dimensional structure thereof are provided herein, as well as methods for crystallizing the same, and using the atomic coordinates of the same to design de novo ligands with high specificity thereto.