Medical concept normalization in clinical trials with drug and disease representation learning

Abstract

Motivation Clinical trials are the essential stage of every drug development program for the treatment to become available to patients. Despite the importance of well-structured clinical trial databases and their tremendous value for drug discovery and development such instances are very rare. Presently large-scale information on clinical trials is stored in clinical trial registers which are relatively structured, but the mappings to external databases of drugs and diseases are increasingly lacking. The precise production of such links would enable us to interrogate richer harmonized datasets for invaluable insights. Results We present a neural approach for medical concept normalization of diseases and drugs. Our two-stage approach is based on Bidirectional Encoder Representations from Transformers (BERT). In the training stage, we optimize the relative similarity of mentions and concept names from a terminology via triplet loss. In the inference stage, we obtain the closest concept name representation in a common embedding space to a given mention representation. We performed a set of experiments on a dataset of abstracts and a real-world dataset of trial records with interventions and conditions mapped to drug and disease terminologies. The latter includes mentions associated with one or more concepts (in-KB) or zero (out-of-KB, nil prediction). Experiments show that our approach significantly outperforms baseline and state-of-the-art architectures. Moreover, we demonstrate that our approach is effective in knowledge transfer from the scientific literature to clinical trial data. Availability We make code and data freely available at hidden\_during\_review\_process Supplementary information Supplementary data are available at Bioinformatics online.

RuCCoN: Clinical Concept Normalization in Russian

“…For training, we have used the publicly available code provided by the authors at https:// github.com/dmis-lab/BioSyn with the following parameters: the number of top candidates k is 20, the mini-batch size is 16, the learning rate is 1e-5, the dense ratio for candidate retrieval is 0.5, the number of epochs is 5. To deal with nil prediction, we apply the strategy from (Miftahutdinov et al, 2021); a mention is out of KB if the nearest candidate is further than a threshold in terms of weighted average of two distances: minimum distance of false positives and maximum distance of true positives, as computed on the train set. Following previous works on entity linking (Suominen et al, 2013;Pradhan et al, 2014;Wright et al, 2019;Phan et al, 2019;Sung et al, 2020;Miftahutdinov et al, 2021;Tutubalina et al, 2020), we use top-k accuracy as the evaluation metric: Acc@k = 1 if the correct CUI is retrieved at rank ≤ k, otherwise Acc@k = 0.…”

“…To deal with nil prediction, we apply the strategy from (Miftahutdinov et al, 2021); a mention is out of KB if the nearest candidate is further than a threshold in terms of weighted average of two distances: minimum distance of false positives and maximum distance of true positives, as computed on the train set. Following previous works on entity linking (Suominen et al, 2013;Pradhan et al, 2014;Wright et al, 2019;Phan et al, 2019;Sung et al, 2020;Miftahutdinov et al, 2021;Tutubalina et al, 2020), we use top-k accuracy as the evaluation metric: Acc@k = 1 if the correct CUI is retrieved at rank ≤ k, otherwise Acc@k = 0. Table 3 shows the Acc@1 and Acc@5 metrics for our test sets.…”

Prediction of Clinical Trials Outcomes Based on Target Choice and Clinical Trial Design with Multi‐Modal Artificial Intelligence

“…The dataset is the property of Insilico Medicine and is commercially available as a part of inClinico platform. All trials were mapped to therapies and conditions by a natural language processing (NLP) pipeline which is based on the state‐of‐the‐art Drug and Disease Interpretation Learning with Biomedical Entity Representation Transformer (DILBERT) 21–23 . This NLP pipeline incorporates two modules: (i) named entity recognition module; and (ii) entity linking module.…”

nablaDFT: Large-Scale Conformational Energy and Hamiltonian Prediction benchmark and dataset

“…One of the key reasons of the success of the Transformer-based models, 31 such as BERT 32 or GPT-3, 33 was the access to a huge training corpus. It has been shown in the domain of medicinal chemistry 34 that degradation in the accuracy from the full dictionary to a 30% of the dictionary is significant for disease linking in clinical trials. Apart from the quality increase, bigger and more diverse datasets are important for models robustness.…”