Evaluation of large language models for discovery of gene set function.

Gene set analysis is a mainstay of functional genomics, but it relies on curated databases of gene functions that are incomplete. Here we evaluate five Large Language Models (LLMs) for their ability to discover the common biological functions represented by a gene set, substantiated by supporting rationale, citations and a confidence assessment. Benchmarking against canonical gene sets from the Gene Ontology, GPT-4 confidently recovered the curated name or a more general concept (73% of cases), while benchmarking against random gene sets correctly yielded zero confidence. Gemini-Pro and Mixtral-Instruct showed ability in naming but were falsely confident for random sets, whereas Llama2-70b had poor performance overall. In gene sets derived from 'omics data, GPT-4 identified novel functions not reported by classical functional enrichment (32% of cases), which independent review indicated were largely verifiable and not hallucinations. The ability to rapidly synthesize common gene functions positions LLMs as valuable 'omics assistants.

Evaluation of large language models for discovery of gene set function.

Gene set analysis is a mainstay of functional genomics, but it relies on manually curated databases of gene functions that are incomplete and unaware of biological context. Here we evaluate the ability of OpenAI's GPT-4, a Large Language Model (LLM), to develop hypotheses about common gene functions from its embedded biomedical knowledge. We created a GPT-4 pipeline to label gene sets with names that summarize their consensus functions, substantiated by analysis text and citations. Benchmarking against named gene sets in the Gene Ontology, GPT-4 generated very similar names in 50% of cases, while in most remaining cases it recovered the name of a more general concept. In gene sets discovered in 'omics data, GPT-4 names were more informative than gene set enrichment, with supporting statements and citations that largely verified in human review. The ability to rapidly synthesize common gene functions positions LLMs as valuable functional genomics assistants.

AMP-BERT: Prediction of antimicrobial peptide function based on a BERT model.

Antimicrobial resistance is a growing health concern. Antimicrobial peptides (AMPs) disrupt harmful microorganisms by nonspecific mechanisms, making it difficult for microbes to develop resistance. Accordingly, they are promising alternatives to traditional antimicrobial drugs. In this study, we developed an improved AMP classification model, called AMP-BERT. We propose a deep learning model with a fine-tuned didirectional encoder representations from transformers (BERT) architecture designed to extract structural/functional information from input peptides and identify each input as AMP or non-AMP. We compared the performance of our proposed model and other machine/deep learning-based methods. Our model, AMP-BERT, yielded the best prediction results among all models evaluated with our curated external dataset. In addition, we utilized the attention mechanism in BERT to implement an interpretable feature analysis and determine the specific residues in known AMPs that contribute to peptide structure and antimicrobial function. The results show that AMP-BERT can capture the structural properties of peptides for model learning, enabling the prediction of AMPs or non-AMPs from input sequences. AMP-BERT is expected to contribute to the identification of candidate AMPs for functional validation and drug development. The code and dataset for the fine-tuning of AMP-BERT is publicly available at https://github.com/GIST-CSBL/AMP-BERT.

Distributed source localization of epileptiform discharges in juvenile myoclonic epilepsy: Standardized low-resolution brain electromagnetic tomography (sLORETA) Study.

Juvenile myoclonic epilepsy (JME) is a common generalized epilepsy syndrome considered the prototype of idiopathic generalized epilepsy. To date, generalized and focal seizures have been the fundamental concepts for classifying seizure types. In several studies, focal features of JME have been reported predominantly in the frontal lobe. However, results in previous studies are inconsistent. Therefore, we investigated the origin of epileptiform discharges in JME. We performed electroencephalography source localization using a distributed model with standardized low-resolution brain electromagnetic tomography. In 20 patients with JME, standardized low-resolution brain electromagnetic tomography images corresponding to the midpoint of the ascending phase and the negative peak of epileptiform discharges were obtained from a total of 362 electroencephalography epochs (181 epochs at each timepoint). At the ascending phase, the maximal current source density was located in the frontal lobe (58.6%), followed by the parietal (26.5%) and occipital lobes (8.8%). At the negative peak, the maximal current source density was located in the frontal lobe (69.1%), followed by the parietal (11.6%) and occipital lobes (9.4%). In the ascending phase, 41.4% of discharges were located outside the frontal lobe, and 30.9% were in the negative peak. Frontal predominance of epileptiform discharges was observed; however, source localization extending to various cortical regions also was identified. This widespread pattern was more prominent in the ascending phase (P = .038). The study results showed that JME includes widespread cortical regions over the frontal lobe. The current concept of generalized epilepsy and pathophysiology in JME needs further validation.

AI-based prediction of new binding site and virtual screening for the discovery of novel P2X3 receptor antagonists.

Artificial intelligence (AI) has been recognized as a powerful technique that can accelerate drug discovery during the hit compound identification step. However, most simple deep learning models have been used for naive pre-filtering as the prediction result cannot be interpreted. Recently, our group developed a new deep learning model (Highlight on Target Sequence; HoTS) that can predict binding regions in a target protein sequence based on patterns learned from interactions between a target protein sequence and a ligand. In this study, we searched for new binding regions of the P2X3 receptor (P2X3R) using HoTS, and suggested a novel putative binding site of P2X3R by a cavity search on the predicted binding regions. The novel putative binding site was employed to generate pharmacophore features, and combinations of pharmacophore features were validated as queries. Two separate virtual screenings using the optimized pharmacophore query Q12 with docking-based scoring and HoTS-based prediction of ligand interactions enabled the initial selection of the compound library for in vitro screening. The screening of each set of 500 compounds from the two approaches (HoTS interaction prediction and Pharmacophore-LibDock cascade) resulted in the identification of 10 (HoTS-1 - 10) and 6 compounds (PD-1 - 6) with low micromolar IC50 values. Remarkably, the hit rate was 10-fold higher than that from the previous random screening of 8364 compound library, and the chemical structures of all identified hit compounds were distinct from those of known P2X3R antagonists, indicating that novel chemical entities could be developed for P2X3R antagonists by targeting the binding site. Overall, this study suggests the discovery of a novel putative binding site for P2X3R using the AI deep learning protocol along with in silico MD simulation and experimental screening of targeted library compounds to successfully identify 16 unique and novel hit compounds. These results may accelerate the discovery of novel chemical-class drugs for P2X3R antagonists.

BayeshERG: a robust, reliable and interpretable deep learning model for predicting hERG channel blockers.

Unintended inhibition of the human ether-à-go-go-related gene (hERG) ion channel by small molecules leads to severe cardiotoxicity. Thus, hERG channel blockage is a significant concern in the development of new drugs. Several computational models have been developed to predict hERG channel blockage, including deep learning models; however, they lack robustness, reliability and interpretability. Here, we developed a graph-based Bayesian deep learning model for hERG channel blocker prediction, named BayeshERG, which has robust predictive power, high reliability and high resolution of interpretability. First, we applied transfer learning with 300 000 large data in initial pre-training to increase the predictive performance. Second, we implemented a Bayesian neural network with Monte Carlo dropout to calibrate the uncertainty of the prediction. Third, we utilized global multihead attentive pooling to augment the high resolution of structural interpretability for the hERG channel blockers and nonblockers. We conducted both internal and external validations for stringent evaluation; in particular, we benchmarked most of the publicly available hERG channel blocker prediction models. We showed that our proposed model outperformed predictive performance and uncertainty calibration performance. Furthermore, we found that our model learned to focus on the essential substructures of hERG channel blockers via an attention mechanism. Finally, we validated the prediction results of our model by conducting in vitro experiments and confirmed its high validity. In summary, BayeshERG could serve as a versatile tool for discovering hERG channel blockers and helping maximize the possibility of successful drug discovery. The data and source code are available at our GitHub repository (https://github.com/GIST-CSBL/BayeshERG).

Comparative analysis of background EEG activity in juvenile myoclonic epilepsy during valproic acid treatment: a standardized, low-resolution, brain electromagnetic tomography (sLORETA) study.

BACKGROUND: By definition, the background EEG is normal in juvenile myoclonic epilepsy (JME) patients and not accompanied by other developmental and cognitive problems. However, some recent studies using quantitative EEG (qEEG) reported abnormal changes in the background activity. QEEG investigation in patients undergoing anticonvulsant treatment might be a useful approach to explore the electrophysiology and anticonvulsant effects in JME. METHODS: We investigated background EEG activity changes in patients undergoing valproic acid (VPA) treatment using qEEG analysis in a distributed source model. In 17 children with JME, non-parametric statistical analysis using standardized low-resolution brain electromagnetic tomography was performed to compare the current density distribution of four frequency bands (delta, theta, alpha, and beta) between untreated and treated conditions. RESULTS: VPA reduced background EEG activity in the low-frequency (delta-theta) bands across the frontal, parieto-occipital, and limbic lobes (threshold log-F-ratio = ±1.414, p < 0.05; threshold log-F-ratio= ±1.465, p < 0.01). In the delta band, comparative analysis revealed significant current density differences in the occipital, parietal, and limbic lobes. In the theta band, the analysis revealed significant differences in the frontal, occipital, and limbic lobes. The maximal difference was found in the delta band in the cuneus of the left occipital lobe (log-F-ratio = -1.840) and the theta band in the medial frontal gyrus of the left frontal lobe (log-F-ratio = -1.610). CONCLUSIONS: This study demonstrated the anticonvulsant effects on the neural networks involved in JME. In addition, these findings suggested the focal features and the possibility of functional deficits in patients with JME.

Sequence-based prediction of protein binding regions and drug-target interactions.

Identifying drug-target interactions (DTIs) is important for drug discovery. However, searching all drug-target spaces poses a major bottleneck. Therefore, recently many deep learning models have been proposed to address this problem. However, the developers of these deep learning models have neglected interpretability in model construction, which is closely related to a model's performance. We hypothesized that training a model to predict important regions on a protein sequence would increase DTI prediction performance and provide a more interpretable model. Consequently, we constructed a deep learning model, named Highlights on Target Sequences (HoTS), which predicts binding regions (BRs) between a protein sequence and a drug ligand, as well as DTIs between them. To train the model, we collected complexes of protein-ligand interactions and protein sequences of binding sites and pretrained the model to predict BRs for a given protein sequence-ligand pair via object detection employing transformers. After pretraining the BR prediction, we trained the model to predict DTIs from a compound token designed to assign attention to BRs. We confirmed that training the BRs prediction model indeed improved the DTI prediction performance. The proposed HoTS model showed good performance in BR prediction on independent test datasets even though it does not use 3D structure information in its prediction. Furthermore, the HoTS model achieved the best performance in DTI prediction on test datasets. Additional analysis confirmed the appropriate attention for BRs and the importance of transformers in BR and DTI prediction. The source code is available on GitHub ( https://github.com/GIST-CSBL/HoTS ).

Lowest threshold lasing modes localized on marginally unstable periodic orbits in a semiconductor microcavity laser.

The lowest threshold lasing mode in a rounded D-shape microcavity is theoretically analyzed and experimentally demonstrated. To identify the lowest threshold lasing mode, we investigate threshold conditions of different periodic orbits by considering the linear gain condition due to the effective pumping region and total loss consisting of internal and scattering losses in ray dynamics. We compare the ray dynamical result with resonance mode analysis, including gain and loss. We find that the resonance modes localized on the pentagonal marginally unstable periodic orbit have the lowest threshold in our fabrication configuration. Our findings are verified by obtaining the path lengths and far-field patterns of lasing modes.

Lasing of scarred mode near above threshold in a semiconductor microcavity laser.

We study a lasing of mode groups in a fully chaotic rounded D-shape InGaAsP semiconductor microcavity laser when an electrode is smaller than a cavity (inward gap). Although there are numerous unstable periodic orbits supporting resonances, a mode group localized on period-5 unstable periodic orbit is more competitive than the others for our laser configuration of the inward gap. By means of theoretical and numerical analyses with ray and wave dynamics, we show that the analyses well agree with our experimental results.

Artificial Intelligence in Drug Discovery: A Comprehensive Review of Data-driven and Machine Learning Approaches.

As expenditure on drug development increases exponentially, the overall drug discovery process requires a sustainable revolution. Since artificial intelligence (AI) is leading the fourth industrial revolution, AI can be considered as a viable solution for unstable drug research and development. Generally, AI is applied to fields with sufficient data such as computer vision and natural language processing, but there are many efforts to revolutionize the existing drug discovery process by applying AI. This review provides a comprehensive, organized summary of the recent research trends in AI-guided drug discovery process including target identification, hit identification, ADMET prediction, lead optimization, and drug repositioning. The main data sources in each field are also summarized in this review. In addition, an in-depth analysis of the remaining challenges and limitations will be provided, and proposals for promising future directions in each of the aforementioned areas.

DeepConv-DTI: Prediction of drug-target interactions via deep learning with convolution on protein sequences.

Identification of drug-target interactions (DTIs) plays a key role in drug discovery. The high cost and labor-intensive nature of in vitro and in vivo experiments have highlighted the importance of in silico-based DTI prediction approaches. In several computational models, conventional protein descriptors have been shown to not be sufficiently informative to predict accurate DTIs. Thus, in this study, we propose a deep learning based DTI prediction model capturing local residue patterns of proteins participating in DTIs. When we employ a convolutional neural network (CNN) on raw protein sequences, we perform convolution on various lengths of amino acids subsequences to capture local residue patterns of generalized protein classes. We train our model with large-scale DTI information and demonstrate the performance of the proposed model using an independent dataset that is not seen during the training phase. As a result, our model performs better than previous protein descriptor-based models. Also, our model performs better than the recently developed deep learning models for massive prediction of DTIs. By examining pooled convolution results, we confirmed that our model can detect binding sites of proteins for DTIs. In conclusion, our prediction model for detecting local residue patterns of target proteins successfully enriches the protein features of a raw protein sequence, yielding better prediction results than previous approaches. Our code is available at https://github.com/GIST-CSBL/DeepConv-DTI.

Source localization of epileptiform discharges in childhood absence epilepsy using a distributed source model: a standardized, low-resolution, brain electromagnetic tomography (sLORETA) study.

Localizing the source of epileptiform discharges in generalized epilepsy has been controversial for the past few decades. Recent neuroimaging studies have shown that epileptiform discharges in generalized epilepsy can be localized to a particular region. Childhood absence epilepsy (CAE) is the most common generalized epilepsy in childhood and is considered the prototype of idiopathic generalized epilepsy (IGE). To better understand electrophysiological changes and their development in CAE, we investigated the origin of epileptiform discharges. We performed distributed source localization with standardized, low-resolution, brain electromagnetic tomography (sLORETA). In 16 children with CAE, sLORETA images corresponding to the midpoint of the ascending phase and the negative peak of the spike were obtained from a total of 242 EEG epochs (121 epochs at each timepoint). Maximal current source density (CSD) was mostly located in the frontal lobe (69.4%). At the gyral level, maximal CSD was most commonly in the superior frontal gyrus (39.3%) followed by the middle frontal gyrus (14.0%) and medial frontal gyrus (8.7%). At the hemisphere level, maximal CSD was dominant in the right cerebral hemisphere (63.6%). These results were consistent at the midpoint of the ascending phase and the negative peak of the spike. Our results demonstrated that the major source of epileptiform discharges in CAE was the frontal lobe. These results suggest that the frontal lobe is involved in generating CAE. This finding is consistent with recent studies that have suggested selective cortical involvement, especially in the frontal regions, in IGE.

Changes in background electroencephalographic activity in benign childhood epilepsy with centrotemporal spikes after oxcarbazepine treatment: a standardized low-resolution brain electromagnetic tomography (sLORETA) study.

BACKGROUND: Several neuroimaging studies have reported neurophysiological alterations in patients with benign childhood epilepsy with centrotemporal spikes (BCECTS). However, reported outcomes have been inconsistent, and the progression of these changes in the brain remains unresolved. Moreover, background electroencephalography (EEG) in cases of BCECTS has not been performed often. METHODS: We investigated background EEG activity changes after six months of oxcarbazepine treatment to better understand the neurophysiological alterations and progression that occur in BCECTS. In 18 children with BCECTS, non-parametric statistical analyses using standardized low resolution brain electromagnetic tomography (sLORETA) were performed to compare the current density distribution of four frequency bands (delta, theta, alpha, and beta) between untreated and treated conditions. RESULTS: Background EEG activity for the delta frequency band was significantly decreased in the fronto-temporal and limbic regions of the left hemisphere after oxcarbazepine treatment (threshold log-F-ratio = ±2.729, P < 0.01). The maximum current density difference was found in the parahippocampal gyrus of the left limbic lobe (Montreal Neurological Institute coordinate [x, y, z = 25, - 20, - 10], Brodmann area 28) (log-F-ratio = 3.081, P < 0.01). CONCLUSIONS: Our results indicate the involvement of the fronto-temporal and limbic cortices in BCECTS, and limbic lobe involvement, including the parahippocampal gyrus, was noted. In addition to evidence of the involvement of the fronto-temporal and limbic cortices in BCECTS, this study also found that an antiepileptic drug could reduce the delta frequency activity of the background EEG in these regions.

Extremely high Q and unidirectional laser emission due to combination of the Kolmogorov-Arnold-Moser barrier and the chaotic sea in a dielectric microdisk.

Emission characteristics of an oval-shaped microcavity laser are studied. In experiments, modes localized on periodic orbits emit unidirectionally with a narrow in-plane divergence angle of about 12 deg. The origin of high directionality is elucidated by means of classical ray dynamics. Wave calculations show that the Q-factors of the resonances are higher than 108. We explain this extraordinary high Q-factor in relation with a dynamical barrier region where Kolmogorov-Arnold-Moser curves significantly obstruct leakages of resonances.

Identification of drug-target interaction by a random walk with restart method on an interactome network.

BACKGROUND: Identification of drug-target interactions acts as a key role in drug discovery. However, identifying drug-target interactions via in-vitro, in-vivo experiments are very laborious, time-consuming. Thus, predicting drug-target interactions by using computational approaches is a good alternative. In recent studies, many feature-based and similarity-based machine learning approaches have shown promising results in drug-target interaction predictions. A previous study showed that accounting connectivity information of drug-drug and protein-protein interactions increase performances of prediction by the concept of 'guilt-by-association'. However, the approach that only considers directly connected nodes often misses the information that could be derived from distance nodes. Therefore, in this study, we yield global network topology information by using a random walk with restart algorithm and apply the global topology information to the prediction model. RESULTS: As a result, our prediction model demonstrates increased prediction performance compare to the 'guilt-by-association' approach (AUC 0.89 and 0.67 in the training and independent test, respectively). In addition, we show how weighted features by a random walk with restart yields better performances than original features. Also, we confirmed that drugs and proteins that have high-degree of connectivity on the interactome network yield better performance in our model. CONCLUSIONS: The prediction models with weighted features by considering global network topology increased the prediction performances both in the training and testing compared to non-weighted models and previous a 'guilt-by-association method'. In conclusion, global network topology information on protein-protein interaction and drug-drug interaction effects to the prediction performance of drug-target interactions.

Comparative analysis of background EEG activity in childhood absence epilepsy during valproate treatment: a standardized, low-resolution, brain electromagnetic tomography (sLORETA) study.

Valproate (VPA) is an antiepileptic drug (AED) used for initial monotherapy in treating childhood absence epilepsy (CAE). EEG might be an alternative approach to explore the effects of AEDs on the central nervous system. We performed a comparative analysis of background EEG activity during VPA treatment by using standardized, low-resolution, brain electromagnetic tomography (sLORETA) to explore the effect of VPA in patients with CAE. In 17 children with CAE, non-parametric statistical analyses using sLORETA were performed to compare the current density distribution of four frequency bands (delta, theta, alpha, and beta) between the untreated and treated condition. Maximum differences in current density were found in the left inferior frontal gyrus for the delta frequency band (log-F-ratio = -1.390, P > 0.05), the left medial frontal gyrus for the theta frequency band (log-F-ratio = -0.940, P > 0.05), the left inferior frontal gyrus for the alpha frequency band (log-F-ratio = -0.590, P > 0.05), and the left anterior cingulate for the beta frequency band (log-F-ratio = -1.318, P > 0.05). However, none of these differences were significant (threshold log-F-ratio = ±1.888, P < 0.01; threshold log-F-ratio = ±1.722, P < 0.05). Because EEG background is accepted as normal in CAE, VPA would not be expected to significantly change abnormal thalamocortical oscillations on a normal EEG background. Therefore, our results agree with currently accepted concepts but are not consistent with findings in some previous studies.

Chirality of a resonance in the absence of backscatterings.

Chirality of a resonance localized on an islands chain is studied in a deformed Reuleaux triangular-shaped microcavity, where clockwise and counter clockwise traveling rays are classically separated. A resonance localized on a period-5 islands chain exhibits chiral emission due to the asymmetric cavity shape. Chirality is experimentally proved in a InGaAsP multi-quantum-well semiconductor laser by showing that the experimental emission characteristics well coincide with the wave dynamical ones.

Distributed source localization of interictal spikes in benign childhood epilepsy with centrotemporal spikes: A standardized low-resolution brain electromagnetic tomography (sLORETA) study.

Benign childhood epilepsy with centrotemporal spikes (BCECTS), also known as Rolandic epilepsy, is the most common benign childhood epilepsy. Centrotemporal spikes are characteristic findings on electroencephalography (EEG). Though the condition is considered benign, many studies have reported some degree of neuropsychological impairment in individuals with BCECTS. There is also growing evidence from neuroimaging studies that BCECTS may affect a larger portion of the brain than originally thought. We performed distributed source localization analysis of interictal spikes in BCECTS. Current-source density (CSD) of the maximal negative peak of the interictal spikes averaged from each of 20 EEG epochs in 11 patients with BCECTS was measured using standardized low-resolution brain electromagnetic tomography (sLORETA). Rolandic area was included in the distribution of the CSD in all of the patients. The significant CSD and its maximal point were distributed in multiple cortical regions over the Rolandic area. It is suggested that the widespread cortical distribution of interictal spikes seen in this study may be associated with atypical presentation and a variety of comorbidities of BCECTS. Our results imply that BCECTS represents a deviation from normal development during a critical period of brain maturation and that children with BECTS might be more likely to need special medical attention.

Unidirectional emission from a cardioid-shaped microcavity laser.

We find unidirectional emission in a cardioid-shaped microcavity laser. When a deformation parameter is well adjusted, rays starting around a period-5 unstable periodic orbit emit unidirectionally. To confirm the emission direction, we fabricate a laser by using an InGaAsP semiconductor and investigate emission characteristics. When the laser is excited by current injection with a dc current, resonances localized on the period-5 unstable periodic orbit emit unidirectionally.