Artificial intelligence in the life sciences最新文献

Proximity-inducing compounds (PICs) are an emergent drug technology through which a protein of interest (POI), often a drug target, is brought into the vicinity of a second protein which modifies the POI’s function, abundance or localisation, giving rise to a therapeutic effect. One of the best-known examples for such compounds are heterobifunctional molecules known as proteolysis targeting chimeras (PROTACs). PROTACs reduce the abundance of the target protein by establishing proximity to an E3 ligase which labels the protein for degradation via the ubiquitin-proteasomal pathway. Design of PROTACs in silico requires the computational prediction of the ternary complex consisting of POI, PROTAC molecule, and the E3 ligase.

We present a novel machine learning-based method for predicting PROTAC-mediated ternary complex structures using Bayesian optimization. We show how a fitness score combining an estimation of protein-protein interactions with PROTAC conformation energy calculations enables the sample-efficient exploration of candidate structures. Furthermore, our method presents two novel scores for filtering and reranking which take PROTAC stability (Autodock-Vina based PROTAC stability score) and protein interaction restraints (the TCP-AIR score) into account. We evaluate our method using DockQ scores on a number of available ternary complex structures (including previously unevaluated cases) and demonstrate that even with a clustering that requires members to have a high similarity, i.e., with smaller clusters, we can assign high ranks to those clusters that contain poses close to the experimentally determined native structure of the ternary complexes. We also demonstrate the resultant improved yield of near-native poses³ in these clusters.

Microplates are indispensable in large-scale biomedical experiments but the physical location of samples and controls on the microplate can significantly affect the resulting data and quality metric values. We introduce a new method based on constraint programming for designing microplate layouts that reduces unwanted bias and limits the impact of batch effects after error correction and normalisation. We demonstrate that our method applied to dose-response experiments leads to more accurate regression curves and lower errors when estimating ${\text{IC}}_{50}$/${\text{EC}}_{50}$, and for drug screening leads to increased precision, when compared to random layouts. It also reduces the risk of inflated scores from common microplate quality assessment metrics such as $Z^{\prime }$ factor and SSMD. We make our method available via a suite of tools (PLAID) including a reference constraint model, a web application, and Python notebooks to evaluate and compare designs when planning microplate experiments.

Deep Learning techniques have significant advantages for mass spectral classification, such as parallelized signal correction and feature extraction. Deep Metric Learning models combine Metric Learning to determine the degree of similarity or difference between a set of mass spectra with the generalization power of Deep Learning to improve feature extraction even further. The two most popular of these models combine multiple neural networks with identical architectures and are commonly called Siamese (SNN) and Triplet Neural Networks (TNN). Herein, using both SNNs and TNNs, we intended to taxonomically categorize two sets of previously-validated mass spectra that corresponded to 30 species of Neotropical arthropods in the Culicidae and Ixodidae families, some of which are disease vectors. The effectiveness of SNNs and TNNs to correctly classify 826 spectra from 12 mosquito species and 310 spectra from 18 species of hard ticks was highly effective, with both algorithms performing with minimal average loss during cross-validation. SNNs produced accuracy rates for ticks and mosquitoes of 91.22% and 94.46%, respectively, while accuracy rates of 93% and 99% were obtained with TNNs. Our results indicate that Deep Metric Learning is a practical machine learning tool for quickly and precisely classifying MALDI-TOF-generated mass spectra of Neotropical and public-health-relevant arthropod species.

In a drug discovery setting, pharmaceutical companies own substantial but confidential datasets. The MELLODDY project developed a privacy-preserving federated machine learning solution and deployed it at an unprecedented scale. Each partner built models for their own private assays that benefitted from a shared representation. Established predictive performance metrics such as AUC ROC or AUC PR are constrained to unseen labeled chemical space and cannot gage performance gains in unlabeled chemical space. Federated learning indirectly extends labeled space, but in a privacy-preserving context, a partner cannot use this label extension for performance assessment. Metrics that estimate uncertainty on a prediction can be calculated even where no label is known. Practically, the chemical space covered with predictions above an uncertainty threshold, reflects the applicability domain of a model. After establishing a link to established performance metrics, we propose the efficiency from the conformal prediction framework (‘conformal efficiency’) as a proxy to the applicability domain size. A documented extension of the applicability domain would qualify as a tangible benefit from federated learning. In interim assessments, MELLODDY partners reported a median increase in conformal efficiency of the federated over the single-partner model of 5.5% (with increases up to 9.7%). Subject to distributional conditions, that efficiency increase can be directly interpreted as the expected increase in conformal i.e. low uncertainty predictions. In conclusion, we present the first indication that privacy-preserving federated machine learning across massive drug-discovery datasets from ten pharma partners indeed extends the applicability domain of property prediction models.

Patents play a crucial role in the drug discovery process by providing legal protection for discoveries and incentivising investments in research and development. By identifying patterns within patent data resources, researchers can gain insight into the market trends and priorities of the pharmaceutical and biotechnology industries, as well as provide additional perspectives on more fundamental aspects such as the emergence of potential new drug targets. In this paper, we used the patent enrichment tool, PEMT, to extract, integrate, and analyse patent literature for rare diseases (RD) and Alzheimer's disease (AD). This is followed by a systematic review of the underlying patent landscape to decipher trends and applications in patents for these diseases. To do so, we discuss prominent organisations involved in drug discovery research in AD and RD. This allows us to gain an understanding of the importance of AD and RD from specific organisational (pharmaceutical or university) perspectives. Next, we analyse the historical focus of patents in relation to individual therapeutic targets and correlate them with market scenarios allowing the identification of prominent targets for a disease. Lastly, we identified drug repurposing activities within the two diseases with the help of patents. This resulted in identifying existing repurposed drugs and novel potential therapeutic approaches applicable to the indication areas. The study demonstrates the expanded applicability of patent documents from legal to drug discovery, design, and research, thus, providing a valuable resource for future drug discovery efforts. Moreover, this study is an attempt towards understanding the importance of data underlying patent documents and raising the need for preparing the data for machine learning-based applications.

Single cell RNA sequencing (scRNA-seq) technologies provide researchers with an unprecedented opportunity to exploit cell heterogeneity. For example, the sequenced cells belong to various cell lineages, which may have different cell fates in stem and progenitor cells. Those cells may differentiate into various mature cell types in a cell differentiation process. To trace the behavior of cell differentiation, researchers reconstruct cell lineages and predict cell fates by ordering cells chronologically into a trajectory with a pseudo-time. However, in scRNA-seq experiments, there are no cell-to-cell correspondences along with the time to reconstruct the cell lineages, which creates a significant challenge for cell lineage tracing and cell fate prediction. Therefore, methods that can accurately reconstruct the dynamic cell lineages and predict cell fates are highly desirable.

In this article, we develop an innovative machine-learning framework called Cell Smoothing Transformation (CellST) to elucidate the dynamic cell fate paths and construct gene networks in cell differentiation processes. Unlike the existing methods that construct one single bulk cell trajectory, CellST builds cell trajectories and tracks behaviors for each individual cell. Additionally, CellST can predict cell fates even for less frequent cell types. Based on the individual cell fate trajectories, CellST can further construct dynamic gene networks to model gene-gene relationships along the cell differentiation process and discover critical genes that potentially regulate cells into various mature cell types.

Natural products are attractive for drug discovery applications because of their distinctive chemical structures, such as an overall large fraction of sp³ carbon atoms, chiral centers (both features associated with structural complexity), large chemical scaffolds, and diversity of functional groups. Furthermore, natural products are used in de novo design and have inspired the development of pseudo-natural products using generative models. Public databases such as the Collection of Open NatUral ProdUcTs and the Universal Natural Product database (UNPD) are rich sources of structures to be used in generative models and other applications. In this work, we report the selection and characterization of the most diverse compounds of natural products from the UNPD using the MaxMin algorithm. The subsets generated with 14,994, 7,497, and 4,998 compounds are publicly available at https://github.com/DIFACQUIM/Natural-products-subsets-generation. We anticipate that the subsets will be particularly useful in building generative models based on natural products by research groups, particularly those with limited access to extensive supercomputer resources.

Ligand-based virtual screening plays an important role for cases in which protein structures are not available. Among ligand-based methods, accurate and fast prediction of protein-ligand binding affinity is crucial for reducing computational cost and exploring the chemical search space efficiently. Here we proposed a CNN-based method, termed as L3D-PLS for building the quantitative structure-activity relationships without target structures. In L3D-PLS, a CNN module was designed for extracting the key interaction features from the grids around aligned ligands, and a partial least square (PLS) model fits the binding affinity with the extracted features of the pre-trained CNN module. In 30 publicly available pre-aligned molecular datasets, L3D-PLS outperformed the traditional CoMFA method. This results highlight that L3D-PLS can be useful for lead optimization based on small datasets which is often true in drug discovery compaign.

The mechanism of action (MoA) of a compound describes the biological interaction through which it produces a pharmacological effect. Multiple data sources can be used for the purpose of predicting MoA, including compound structural information, and various assays, such as those based on cell morphology, transcriptomics and metabolomics. In the present study we explored the benefits and potential additive/synergistic effects of combining structural information, in the form of Morgan fingerprints, and morphological information, in the form of five-channel Cell Painting image data. For a set of 10 well represented MoA classes, we compared the performance of deep learning models trained on the two datasets separately versus a model trained on both datasets simultaneously. On a held-out test set we obtained a macro-averaged F1 score of 0.58 when training on only the structural data, 0.81 when training on only the image data, and 0.92 when training on both together. Thus indicating clear additive/synergistic effects and highlighting the benefit of integrating multiple data sources for MoA prediction.