Using BERT to identify drug-target interactions from whole PubMed

Background: Drug-target interactions (DTIs) are critical for drug repurposing and elucidation of drug mechanisms, and are manually curated by large databases, such as ChEMBL, BindingDB, DrugBank and DrugTargetCommons. However, the number of articles providing this data (~0.1 million) likely constitutes only a fraction of all articles on PubMed that contain experimentally determined DTIs. Finding such articles and extracting the experimental information is a challenging task, and there is a pressing need for systematic approaches to assist the curation of DTIs. To this end, we propose Bidirectional Encoder Representations from Transformers (BERT) to identify such articles. Because DTI data intimately depends on the type of assays used to generate it, we also aimed to incorporate functions to predict the assay format. Results: Our novel method identified ~2.1 million articles (along with drug and protein information) that are not previously included in public DTI databases. Using 10-fold cross-validation, we obtained ~99% accuracy for identifying articles containing quantitative drug-target profiles. The accuracy for the prediction of assay format is ~90%, which leaves room for improvement in future studies. Conclusion: The BERT model in this study is robust and the proposed pipeline can be used to identify previously overlooked articles containing quantitative DTIs. Overall, our method provides a significant advancement in machine-assisted DTI extraction and curation. We expect it to be a useful addition to drug mechanism discovery and repurposing.

Capture the Tag: Mitigation of Biotin Interference in ELISA and Automated Immunoassays by Pre-Conjugating Biotinylated Antibodies to the Streptavidin Surface

Introduction Streptavidin to biotin binding is one of the strongest non-covalent interactions in nature, and is therefore, successfully incorporated into many immunoassays to facilitate antibody capture. Biotin-streptavidin coupling assays are susceptible to interference from free biotin in patient specimens, which may falsely decrease or increase the result depending on assay format. Currently, biotin-stripping methods capture free biotin in patient specimens by pre-incubation with streptavidin-coated microparticles. However, this approach increases turnaround time, test cost, and testing steps, which increases the chance of error. Our objective was to determine if pre-conjugating biotinylated antibodies to the assay’s streptavidin solid surface before adding patient specimen could mitigate biotin interference in ELISA and automated sandwich immunoassays. Methods We performed this study on 3 different ELISAs (CYFRA-21, NSE, S100B) and one automated assay (thyroglobulin); all are a sandwich immunoassay format. Serum pools were spiked with biotin (25 - 1000 µg/L) or PBS control. Manufacturer protocols were followed in both the ELISAs and automated assay to evaluate baseline concentration-dependent biotin interference. Mitigation of biotin interference by pre-incubation was then evaluated in the ELISAs by adding biotinylated antibody to the streptavidin-coated wells 0, 10, 15, or 60 min before adding biotin- or PBS-spiked serum specimens. For the automated assay, streptavidin-coated beads and biotinylated antibody were removed from the reagent cartridge, mixed, and incubated 4.5 hours. The mixture containing biotin-tagged antibody bound to streptavidin beads was added back to the cartridge and placed on the analyzer to evaluate spiked specimens. Lastly, we compared the pre-incubation method to a standard biotin-stripping protocol in the ELISAs to compare the effectiveness of mitigating biotin interference. Results We observed biotin interference across the three ELISAs, where 400 µg/L biotin spiked in serum pools reduced analyte detection to between 10 – 15% of the total activity using the standard assay format. Our time-course studies showed 84 – 95% recovery of the total activity when the biotinylated antibody was pre-incubated in the streptavidin-coated wells for 1 hour prior to addition of serum specimens, compared to 69 – 99% by a standard biotin stripping protocol. We extended this concept to the automated immunoassay where at 1000 µg/L biotin in the specimen, only 9 ±0.01% of the total analyte was measured by the conventional method. However, pre-mixing biotinylated antibody and streptavidin beads resulted in 97 ±0.01% of the total analyte recovery in the presence of 1000 µg/L biotin. Conclusion We have demonstrated that pre-conjugating the biotin antibody to streptavidin is as effective as biotin-stripping methods to avoid biotin interference in sandwich immunoassays that utilize the biotin-streptavidin system, with the additional benefit of optimizing turnaround times, cost, and labor. A simple change in manufacturer assay design could make immunoassays more robust against biotin interference in patient samples.

Using BERT to identify drug-target interactions from whole PubMed

ABSTRACTBackgroundDrug-target interactions (DTIs) are critical for drug repurposing and elucidation of drug mechanisms, and they are collected in large databases, such as ChEMBL, BindingDB, DrugBank and DrugTargetCommons. However, the number of studies providing this data (~0.1 million) likely constitutes only a fraction of all studies on PubMed that contain experimental DTI data. Finding such studies and extracting the experimental information is a challenging task, and there is a pressing need for machine learning for the extraction and curation of DTIs. To this end, we developed new text mining document classifiers based on the Bidirectional Encoder Representations from Transformers (BERT) algorithm. Because DTI data intimately depends on the type of assays used to generate it, we also aimed to incorporate functions to predict the assay format.ResultsOur novel method identified and extracted DTIs from 2.1 million studies not previously included in public DTI databases. Using 10-fold cross-validation, we obtained ~99% accuracy for identifying studies containing drug-target pairs. The accuracy for the prediction of assay format is ~90%, which leaves room for improvement in future studies.ConclusionThe BERT model in this study is robust and the proposed pipeline can be used to identify new and previously overlooked studies containing DTIs and automatically extract the DTI data points. The tabular output facilitates validation of the extracted data and assay format information. Overall, our method provides a significant advancement in machine-assisted DTI extraction and curation. We expect it to be a useful addition to drug mechanism discovery and repurposing.

Novel methodology comparing two assay formats for the assessment of immunogenicity from clinical trial samples

Aim: We present a novel methodology to compare results between distinct immunogenicity assays, performed by two laboratories, for the same biotherapeutic. Materials & methods: Human serum pools from clinical trials were generated to provide representative immunogenicity titers. Pools were evaluated at two laboratories in a blinded fashion to assess the effect of assay format and laboratory change on clinical interpretation of immunogenicity results. Results: The laboratories validated two different assay formats and demonstrated comparable sensitivity and drug tolerance. Overall, the comparisons in assay format and laboratory ensured a comparable ability to detect treatment-emergent antidrug antibodies for a biotherapeutic. Conclusion: We have established an approach, using pooling of patient samples, that allows for the interlaboratory comparisons without creating duplicative results.

The importance of quality critical reagents for the entire developmental lifecycle of a biopharmaceutical: a pharmacokinetic case study

Background: High-quality critical reagents are essential to the successful support of biotherapeutic drug development regardless of the analytical platform used for support. The lack of such a reagent, early in the development lifecycle of a biotherapeutic can have detrimental impact on resource and translation of data across development phases. Results: Here, a pharmacokinetic assay case study is shared that illustrates what can occur when there is a lack of a reproducible and sustainable critical reagent early in the development lifecycle of a biotherapeutic. Various assay formats and critical reagents, as well as reagents generation programs, were initiated to find a reagent and assay format which was fit for purpose. Conclusions: Identification of appropriate critical reagents early in the development lifecycle of a biotherapeutic as advantageous.

Optimization of Beer Brewing by Monitoring α-Amylase and β-Amylase Activities during Mashing

(1) Background: In the current highly competitive brewing industry, most breweries may benefit from a reduction in mashing time. In this study, a novel enzymatic assay format was used to investigate the activities of α-amylase and β-amylase during different mashing profiles, with the aim to use it as a tool for optimizing the production time of an existing industrial mashing process; (2) Methods: Lab-scale mashings with eight different time-temperature programs and two different pilot brews were analyzed in terms of enzymatic activity, sugar composition, alcohol by volume in the final beer, FAN and others; (3) Results: A 20-min reduction (out of an original 73-min mashing program) was achieved by selecting a temperature profile which maintained a higher enzymatic activity than the original, without affecting the wort sugar composition and fermentability, or the ethanol concentration and foam stability of the final beer. (4) Conclusions: A method is presented which can be used by breweries to optimize their mashing profiles based on monitoring α-amylase and β-amylase activities.

Enhanced Performance of DNA Methylation Markers by Simultaneous Measurement of Sense and Antisense DNA Strands after Cytosine Conversion

Background Most existing DNA methylation-based methods for detection of circulating tumor DNA (ctDNA) are based on conversion of unmethylated cytosines to uracil. After conversion, the 2 DNA strands are no longer complementary; therefore, targeting only 1 DNA strand merely utilizes half of the available input DNA. We investigated whether the sensitivity of methylation-based ctDNA detection strategies could be increased by targeting both DNA strands after bisulfite conversion. Methods Dual-strand digital PCR assays were designed for the 3 colorectal cancer (CRC)–specific methylation markers KCNQ5, C9orf50, and CLIP4 and compared with previously reported single-strand assays. Performance was tested in tumor and leukocyte DNA, and the ability to detect ctDNA was investigated in plasma from 43 patients with CRC stages I to IV and 42 colonoscopy-confirmed healthy controls. Results Dual-strand assays quantified close to 100% of methylated control DNA input, whereas single-strand assays quantified approximately 50%. Furthermore, dual-strand assays showed a 2-fold increase in the number of methylated DNA copies detected when applied to DNA purified from tumor tissue and plasma from CRC patients. When the results of the 3 DNA methylation markers were combined into a ctDNA detection test and applied to plasma, the dual-strand assay format detected 86% of the cancers compared with 74% for the single-strand assay format. The specificity was 100% for both the dual- and single-strand test formats. Conclusion Dual-strand assays enabled more sensitive detection of methylated ctDNA than single-strand assays.