On the use of real-world datasets for reaction yield prediction.
Ontology highlight
ABSTRACT:
SUBMITTER: Saebi M
PROVIDER: S-EPMC10189898 | biostudies-literature | 2023 May
REPOSITORIES: biostudies-literature
Publications
On the use of real-world datasets for reaction yield prediction.
Saebi Mandana M Nan Bozhao B Herr John E JE Wahlers Jessica J Guo Zhichun Z Zurański Andrzej M AM Kogej Thierry T Norrby Per-Ola PO Doyle Abigail G AG Chawla Nitesh V NV Wiest Olaf O
Chemical science 20230313 19
The lack of publicly available, large, and unbiased datasets is a key bottleneck for the application of machine learning (ML) methods in synthetic chemistry. Data from electronic laboratory notebooks (ELNs) could provide less biased, large datasets, but no such datasets have been made publicly available. The first real-world dataset from the ELNs of a large pharmaceutical company is disclosed and its relationship to high-throughput experimentation (HTE) datasets is described. For chemical yield ...[more]
Similar Datasets
Project description:The chemical reaction yield is an important factor to determine the reaction conditions. Recently, many data-driven models for yield prediction using high-throughput experimentation datasets have been reported. In this study, we propose a neural network architecture based on the chemical graphs of the reaction components to predict the reaction yield. The proposed model is the sequential combination of a message-passing neural network and a transformer encoder (MPNN-Transformer). The reaction components are converted to molecular matrices by the first network, followed by the interplay of the reaction components in the second network after adding the embeddings of the compound roles in the chemical reaction. The predictive ability of the proposed models was compared with state-of-the-art yield prediction models using two high-throughput experimental datasets: the Buchwald-Hartwig cross-coupling (BHC) and Suzuki-Miyaura cross-coupling (SMC) reaction datasets. Overall, the MPNN-Transformer models showed high prediction accuracy for the BHC reaction datasets and some of the extrapolation-oriented SMC reaction datasets. These models also performed well when the training dataset size was relatively large. Furthermore, analyzing the poorly predicted reactions for the BHC reaction dataset revealed a limitation of the data-driven yield prediction approach based on the chemical structural similarity.
| S-EPMC11447720 | biostudies-literature
Project description:There are several important challenges in radiomics research; one of them is feature selection. Since many quantitative features are non-informative, feature selection becomes essential. Feature selection methods have been mixed with filter, wrapper, and embedded methods without a rule of thumb. This study aims to develop a framework for optimal feature selection in radiomics research. We developed the framework that the optimal features were selected to quickly through controlling relevance and redundancy among features. A 'FeatureMap' was generated containing information for each step and used as a platform. Through this framework, we can explore the optimal combination of radiomics features and evaluate the predictive performance using only selected features. We assessed the framework using four real datasets. The FeatureMap generated 6 combinations, with the number of features selected varying for each combination. The predictive models obtained high performances; the highest test area under the curves (AUCs) were 0.792, 0.820, 0.846 and 0.738 in the cross-validation method, respectively. We developed a flexible framework for feature selection methods in radiomics research and assessed its usefulness using various real-world data. Our framework can assist clinicians in efficiently developing predictive models based on radiomics.
| S-EPMC11599926 | biostudies-literature
Project description:Predicting the nature and outcome of reactions using computational methods is a crucial tool to accelerate chemical research. The recent application of deep learning-based learned fingerprints to reaction classification and reaction yield prediction has shown an impressive increase in performance compared to previous methods such as DFT- and structure-based fingerprints. However, learned fingerprints require large training data sets, are inherently biased, and are based on complex deep learning architectures. Here we present the differential reaction fingerprint DRFP. The DRFP algorithm takes a reaction SMILES as an input and creates a binary fingerprint based on the symmetric difference of two sets containing the circular molecular n-grams generated from the molecules listed left and right from the reaction arrow, respectively, without the need for distinguishing between reactants and reagents. We show that DRFP performs better than DFT-based fingerprints in reaction yield prediction and other structure-based fingerprints in reaction classification, reaching the performance of state-of-the-art learned fingerprints in both tasks while being data-independent.
| S-EPMC8996827 | biostudies-literature
Project description:Deep learning (DL)-driven efficient synthesis planning may profoundly transform the paradigm for designing novel pharmaceuticals and materials. However, the progress of many DL-assisted synthesis planning (DASP) algorithms has suffered from the lack of reliable automated pathway evaluation tools. As a critical metric for evaluating chemical reactions, accurate prediction of reaction yields helps improve the practicality of DASP algorithms in the real-world scenarios. Currently, accurately predicting yields of interesting reactions still faces numerous challenges, mainly including the absence of high-quality generic reaction yield datasets and robust generic yield predictors. To compensate for the limitations of high-throughput yield datasets, we curated a generic reaction yield dataset containing 12 reaction categories and rich reaction condition information. Subsequently, by utilizing 2 pretraining tasks based on chemical reaction masked language modeling and contrastive learning, we proposed a powerful bidirectional encoder representations from transformers (BERT)-based reaction yield predictor named Egret. It achieved comparable or even superior performance to the best previous models on 4 benchmark datasets and established state-of-the-art performance on the newly curated dataset. We found that reaction-condition-based contrastive learning enhances the model's sensitivity to reaction conditions, and Egret is capable of capturing subtle differences between reactions involving identical reactants and products but different reaction conditions. Furthermore, we proposed a new scoring function that incorporated Egret into the evaluation of multistep synthesis routes. Test results showed that yield-incorporated scoring facilitated the prioritization of literature-supported high-yield reaction pathways for target molecules. In addition, through meta-learning strategy, we further improved the reliability of the model's prediction for reaction types with limited data and lower data quality. Our results suggest that Egret holds the potential to become an essential component of the next-generation DASP tools.
| S-EPMC10777739 | biostudies-literature
Project description:Urothelial carcinoma (UC) of the urinary bladder has significant challenges in treatment due to its diverse genetic landscape and variable response to systemic therapy. In recent years, patient-derived organoids (PDOs) emerged as a novel tool to model primary tumors with higher resemblance than conventional 2D cell culture approaches. However, the potential of organoids to predict therapy response in a clinical setting remains to be evaluated. This study explores the clinical feasibility of PDOs for pharmacotyping in UC. Initially, we subjected tumor tissue specimens from 50 patients undergoing transurethral resection or radical cystectomy to organoid propagation, of whom 19 (38%) yielded PDOs suitable for drug sensitivity assessment. Notably, whole transcriptome-based analysis indicated that PDOs may show phenotypes distinct from their parental tumor tissue. Pharmacotyping within a clinically relevant timeframe [mean of 35.44 and 55 days for non-muscle invasive bladder cancer (NMIBC) and muscle invasive bladder cancer (MIBC), respectively] was achieved. Drug sensitivity analyses revealed marked differences between NMIBC and MIBC, with MIBC-derived organoids demonstrating higher chemosensitivity toward clinically relevant drugs. A case study correlating organoid response with patient treatment outcome illustrated the complexity of predicting chemotherapy efficacy, especially considering the rapid acquisition of drug resistance. We propose a workflow of prospective organoid-based pharmacotyping in UC, enabling further translational research and integration of this approach into clinical practice.
2024-11-03 | GSE280749 | GEO
Project description:We present comprehensive datasets of Brazilian disasters from January 2003 to February 2021 as well as real-world optimization instances built up from these data. The data were gathered through a series of open available reports obtained from different government and institutional sources. Afterwards, data consolidation and summarization were carried out using Excel and Python. The datasets include 9 types of disaster, such as flash floods, landslides and droughts, and the corresponding number of affected people during an 18-year or a 218-month observation period for 5,402 Brazilian municipalities, totaling more than 65,000 observations. Data on relevant geographical, demographic and socioeconomic aspects of the affected municipalities are also provided. These encompass geographic coordinates, regions, population, income, development indicators, amongst other information. From a statistical point of view, the data on disasters can address a number of applications using both supervised and unsupervised machine learning techniques such as, for time series analysis or other dynamic models using socioeconomic data as explanatory variables, i.e. data on the size of the poor population, income, education and general development. The geographic dataset can be useful for aggregating analyses concerning the various forms of territorial organization and allows for the visualization of data in maps. All the aforementioned data can be also used to devise realistic optimization instances related to diverse humanitarian logistics and/or disaster management problems, such as facility location, location-allocation, vehicle routing, and so forth. In particular, we describe two real-world instances for the location-allocation problem studied in [1]. For that purpose, we partially use the given datasets and included other information such as costs and distances relevant to the optimization model. Although using real-world cases to test optimization approaches is a common and encouraged practice in Operations Research, comprehensive datasets and practical optimization instances, as presented in this article, are rarely described and/or available in the academic literature.
| S-EPMC8931360 | biostudies-literature
Project description:There is growing interest in regulatory use of randomized pragmatic trials and noninterventional real-world (RW) studies of effectiveness and safety, but there is no agreed-on framework for assessing when this type of evidence is sufficiently reliable. Rather than impose a clinical trial-like paradigm on RW evidence, like blinded treatments or complete, source-verified data, the framework for assessing the utility of RW evidence should be grounded in the context of specific study objectives, clinical events that are likely to be detected in routine care, and the extent to which systematic error (bias) is likely to impact effect estimation. Whether treatment is blinded should depend on how well the outcome can be measured objectively. Qualification of a data source should be based on (1) numbers of patients of interest available for study; (2) if "must-have" data are likely to be recorded, and if so, how and where; (3) the accessibility of systematic follow-up data for the time period of interest; and (4) the potential for systematic errors (bias) in data collection and the likely magnitude of any such bias. Accessible data may not be representative of an entire population, but still may provide reliable evidence about the experience of typical patients treated under conditions of conventional care. Similarly, RW data that falls short of optimal length of follow-up or study size may still be useful in terms of its ability to provide evidence for regulators for subgroups of special interest. Developing a framework to qualify RW evidence in the context of a particular study purpose and data asset will enable broader regulatory use of RW data for approval of new molecular entities and label changes. Reliable information about diverse populations and settings should also help us move closer to more affordable, effective health care.
| S-EPMC5944086 | biostudies-other
Project description:The systematic substitution of direct observational data with synthesized data derived from models during the stock assessment process has emerged as a low-cost alternative to direct data collection efforts. What is not widely appreciated, however, is how the use of such synthesized data can overestimate predictive skill when forecasting recruitment is part of the assessment process. Using a global database of stock assessments, we show that Standard Fisheries Models (SFMs) can successfully predict synthesized data based on presumed stock-recruitment relationships, however, they are generally less skillful at predicting observational data that are either raw or minimally filtered (denoised without using explicit stock-recruitment models). Additionally, we find that an equation-free approach that does not presume a specific stock-recruitment relationship is better than SFMs at predicting synthesized data, and moreover it can also predict observational recruitment data very well. Thus, while synthesized datasets are cheaper in the short term, they carry costs that can limit their utility in predicting real world recruitment.
| S-EPMC7181812 | biostudies-literature
Project description:BackgroundThe proportion of days covered (PDC) is used to estimate medication adherence by looking at the proportion of days in which a person has access to the medication, over a given period of interest. This study aimed to adapt the PDC algorithm to allow for plausible assumptions about prescription refill behaviour when applied to data from online pharmacy suppliers.MethodsThree PDC algorithms, the conventional approach (PDC1) and two alternative approaches (PDC2 and PDC3), were used to estimate adherence in a real-world dataset from an online pharmacy. Each algorithm has different denominators and increasing levels of complexity. PDC1, the conventional approach, is the total number of days between first dispensation and a defined end date. PDC2 counts the days until the end of supply date. PDC3 removes from the denominator specifically defined large gaps between refills, which could indicate legitimate reasons for treatment discontinuation. The distribution of the three PDCs across four different follow-up lengths was compared.ResultsThe dataset included people taking ACE inhibitors (n = 65,905), statins (n = 100,362), and/or thyroid hormones (n = 30,637). The proportion of people taking ACE inhibitors with PDC ≥ 0.8 was 50-74% for PDC1, 81-91% for PDC2, and 86-100% for PDC3 with values depending on drug and length of follow-up. Similar ranges were identified in people taking statins and thyroid hormones.ConclusionThese algorithms enable researchers and healthcare providers to assess pharmacy services and individual levels of adherence in real-world databases, particularly in settings where people may switch between different suppliers of medicines, meaning an individual supplier's data may show temporary but legitimate gaps in access to medication. Accurately identifying problems with adherence provides the foundation for opportunities to improve experience, adherence and outcomes and to reduce medicines wastage. Research with people taking medications and prescribers is required to validate the algorithms' assumptions.
| S-EPMC8715592 | biostudies-literature
Project description:Graph neural networks (GNNs) have proven to be effective in the prediction of chemical reaction yields. However, their performance tends to deteriorate when they are trained using an insufficient training dataset in terms of quantity or diversity. A promising solution to alleviate this issue is to pre-train a GNN on a large-scale molecular database. In this study, we investigate the effectiveness of GNN pre-training in chemical reaction yield prediction. We present a novel GNN pre-training method for performance improvement.Given a molecular database consisting of a large number of molecules, we calculate molecular descriptors for each molecule and reduce the dimensionality of these descriptors by applying principal component analysis. We define a pre-text task by assigning a vector of principal component scores as the pseudo-label to each molecule in the database. A GNN is then pre-trained to perform the pre-text task of predicting the pseudo-label for the input molecule. For chemical reaction yield prediction, a prediction model is initialized using the pre-trained GNN and then fine-tuned with the training dataset containing chemical reactions and their yields. We demonstrate the effectiveness of the proposed method through experimental evaluation on benchmark datasets.
| S-EPMC10905905 | biostudies-literature