Insitro, Inc.

Embodiments of the disclosure involve modeling DEL data using factorized molecular representations (e.g., hierarchical mono-synthon and di-synthon building blocks), which capitalizes on the inherent hierarchical structure of these molecules. Using the factorized molecular representations, machine learning models are trained to learn latent binding affinity of compounds for targets and one or more covariates (e.g., load/replicate noise). This leads to improved predictions by the machine learning models in the form of higher enrichment scores, which are well-correlated with compound-target binding affinity.

The present disclosure relates to multifunctional molecules, including molecules according to formula (I): ([(B 1 ) M -D-L 1 ] Y —H 1 ) O -G-(H 2 -[L 2 -E-(B 2 ) K ] W ) P ,  (I) wherein G, H 1 , H 2 , D, E, B 1 , B 2 , M, K, L 1 , L 2 , O, P, Y, and W are defined herein. The present disclosure also relates to methods of preparing and using such multifunctional molecules to identify encoded molecules capable of binding target molecules.

The present disclosure relates generally to providing a cellular time-series imaging, modeling, and analysis platform, and more specifically to acquiring time-series image data and using various machine learning models to model and analyze subcellular particle movements and changes in cellular positional and morphological characteristics using unsupervised embedding generation. The platform can be applied to evaluate various cellular and subcellular processes by generating summary embeddings of time-series image data that enable analysis of dynamic cellular and subcellular processes over time (e.g., the movement of particles within a cell, neurites on developing neurons, etc.) for enhanced identification of differences between cell states (e.g., between sick and healthy cells) and generation of disease models which can be used to analyze the impact of various therapeutic interventions, among other improvements described throughout.

Embodiments of the disclosure include systems and non-transitory computer readable media for analyzing microscopy images for developing machine learning models for disease modeling. Microscopy images are captured from cells of one or more exposure response phenotypes (ERPs) and further used to train machine learning models. Thus, trained machine learning models can distinguish between microscopy images captured from healthy and diseased samples.

The present disclosure relates generally to an autonomous cell imaging and modeling platform, and more specifically to machine-learning techniques for using microscopy imaging data to continuously study live biological cells. The autonomous cell imaging and modeling platform can be applied to evaluate various cellular processes, such as cellular differentiation, optimization of cell culture (e.g., in-plate cytometry), disease modeling, histopathology imaging, and genetic and chemical screening, using a dynamic universal imaging system. In some embodiments, the platform comprises a set of label-free computational imaging techniques, self-supervised learning models, and robotic devices configured in an autonomous imaging system to study positional and morphological characteristics in particular cellular substructures of a cell culture in an efficient and non-destructive manner over time.

The present disclosure relates to precursor molecules of DNA-encoded compounds, and methods of preparing thereof. In some aspects, provided herein are methods of synthesizing DNA-encoded compounds, and libraries thereof, from precursor molecules and positional building blocks.

An exemplary method for determining a sampling protocol for sampling tissue cores for a tissue microarray includes obtaining an initial plurality of tissue cores from an image of a tissue slide; selecting a first subset of the initial plurality of tissue cores based on a first candidate sampling protocol; inputting the first subset of the plurality of tissue cores into a machine learning model; evaluating the first candidate sampling protocol by evaluating a first output of the machine learning model; selecting a second subset of the initial plurality of tissue cores based on a second candidate sampling protocol; inputting the second subset of the plurality of tissue cores into the machine learning model; evaluating the second candidate sampling protocol by evaluating a second output of the machine learning model; and determining the sampling protocol based on the evaluation of the first candidate sampling protocol and the second candidate sampling protocol.

An exemplary method for determining a sampling protocol for sampling tissue cores for a tissue microarray includes obtaining an initial plurality of tissue cores from an image of a tissue slide; selecting a first subset of the initial plurality of tissue cores based on a first candidate sampling protocol; inputting the first subset of the plurality of tissue cores into a machine learning model; evaluating the first candidate sampling protocol by evaluating a first output of the machine learning model; selecting a second subset of the initial plurality of tissue cores based on a second candidate sampling protocol; inputting the second subset of the plurality of tissue cores into the machine learning model; evaluating the second candidate sampling protocol by evaluating a second output of the machine learning model; and determining the sampling protocol based on the evaluation of the first candidate sampling protocol and the second candidate sampling protocol.

Embodiments of the disclosure involve modeling DEL data using factorized molecular representations (e.g., hierarchical mono-synthon and di-synthon building blocks), which capitalizes on the inherent hierarchical structure of these molecules. Using the factorized molecular representations, machine learning models are trained to learn latent binding affinity of compounds for targets and one or more covariates (e.g., load/replicate noise). This leads to improved predictions by the machine learning models in the form of higher enrichment scores, which are well-correlated with compound-target binding affinity.

The present disclosure relates generally to providing a cellular time-series imaging, modeling, and analysis platform, and more specifically to acquiring time-series image data and using various machine learning models to model and analyze subcellular particle movements and changes in cellular positional and morphological characteristics using unsupervised embedding generation. The platform can be applied to evaluate various cellular and subcellular processes by generating summary embeddings of time-series image data that enable analysis of dynamic cellular and subcellular processes over time (e.g., the movement of particles within a cell, neurites on developing neurons, etc.) for enhanced identification of differences between cell states (e.g., between sick and healthy cells) and generation of disease models which can be used to analyze the impact of various therapeutic interventions, among other improvements described throughout.