Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

Deep-Learning Resources for Studying Glycan-Mediated Host-Microbe Interactions.

In Cell host & microbe ; h5-index 102.0

Glycans, the most diverse biopolymer, are shaped by evolutionary pressures stemming from host-microbe interactions. Here, we present machine learning and bioinformatics methods to leverage the evolutionary information present in glycans to gain insights into how pathogens and commensals interact with hosts. By using techniques from natural language processing, we develop deep-learning models for glycans that are trained on a curated dataset of 19,299 unique glycans and can be used to study and predict glycan functions. We show that these models can be utilized to predict glycan immunogenicity and the pathogenicity of bacterial strains, as well as investigate glycan-mediated immune evasion via molecular mimicry. We also develop glycan-alignment methods and use these to analyze virulence-determining glycan motifs in the capsular polysaccharides of bacterial pathogens. These resources enable one to identify and study glycan motifs involved in immunogenicity, pathogenicity, molecular mimicry, and immune evasion, expanding our understanding of host-microbe interactions.

Bojar Daniel, Powers Rani K, Camacho Diogo M, Collins James J


bioinformatics, deep learning, glycans, glycobiology, host-microbe, machine learning

General General

Objective measurement of limb bradykinesia using a marker-less tracking algorithm with 2D-video in PD patients.

In Parkinsonism & related disorders ; h5-index 58.0

BACKGROUND : Quantitative measurement of parkinsonian motor symptoms is crucial in clinical practice and in research. However, the widely used Unified PD Rating Scale (UPDRS) part III is based on a semi-quantitative evaluation with high inter- and intra-rater variability. Sensor-based measurements have been widely studied but are limited for their accessibility.

METHODS : We analyzed 2D-RGB videos recording finger tapping and leg agility tests in 29 PD patients with a marker-less deep-learning based tracking algorithm. The tracking performance was validated with an accelerometer. Four parameters (mean amplitude, mean interpeak interval, amplitude variability and interpeak interval variability) were calculated from the position tracking.

RESULTS : The performance of the video-tracking was in good agreement with the accelerometer-based tracking (Intra-class correlation coefficient > 0.9 for the peak amplitude, and >0.6 for the interpeak interval). The video-tracking successfully captured variable aspects of limb bradykinesia that have a distinct correlation with the general parkinsonian motor symptoms and gait. In the finger-tapping task, the mean amplitude (R = -0.6, p = 2.4 × 10-6), amplitude variability (R = 0.36, p = 0.0092), mean interpeak interval (R = 0.34, p = 0.014), and interpeak interval variability (R = 0.66, p = 1.4 × 10-7) was significantly correlated with the UPDRS scores. In leg agility test, the mean amplitude (R = -0.58, p = 1.7 × 10-5), mean interpeak interval (R = 0.37, p = 0.0088) and interpeak interval variability (R = 0.7, p = 6.2 × 10-8) were significantly correlated with the UPDRS scores, but not with amplitude variability (R = 0.17, p = 0.26). Limb rigidity was significantly correlated with the interpeak interval (R = 0.40, p = 0.0036) and its variability (R = 0.59, p = 4.2 × 10-6) in the leg agility test.

CONCLUSION : The video-based tracking could objectively measure limb bradykinesia in PD patients.

Shin Jung Hwan, Ong Jed Noel, Kim Ryul, Park Sang-Min, Choi Jihyun, Kim Han-Joon, Jeon Beomseok


Deep-learning, Finger tapping test, Leg agility test, Parkinson disease, Video tracking

oncology Oncology

Academics as leaders in the cancer artificial intelligence revolution.

In Cancer ; h5-index 88.0

The successful translation of artificial intelligence (AI) applications into clinical cancer care practice requires guidance by academic cancer researchers and providers who are well poised to step into leadership roles. In this commentary, the authors describe the landscape of the deep learning-based AI innovation boom in cancer research. For progress in applied AI research to continue, 4 essential components must be present: algorithms, data, computational resources, and domain-specific expertise. Each of these components is available to researchers and providers in academic settings; in particular, cancer care domain-specific expertise in academia is superb. Three common pitfalls for deep learning research also are detailed along with a discussion of how the academic oncology research environment is well suited to guard against these challenges. In this rapidly developing field, there are few established standards, and oncology researchers and providers must educate themselves about emerging AI technology to avoid common pitfalls and ensure responsible use.

Kochanny Sara E, Pearson Alexander T


artificial intelligence, deep learning, histology, oncology, precision medicine

Radiology Radiology

Deep Learning Based Radiomics Predicts Response to Chemotherapy in Colorectal Liver Metastases.

In Medical physics ; h5-index 59.0

PURPOSE : To develop and validate a deep learning (DL)-based radiomics model to predict the response to chemotherapy in colorectal liver metastases (CRLM).

METHODS : In this retrospective study, we enrolled 192 patients diagnosed with CRLM who received first-line chemotherapy and were followed by response assessment. Tumor response was identified according to the Response Evaluation Criteria in Solid Tumors (RECIST). Contrast-enhanced multidetector computed tomography (MDCT) images were fed as inputs of the ResNet10-based DL radiomics model, and the possibility of response was predicted as the output. The final combined DL radiomics model was constructed by integrating the response related clinical factors and the developed DL radiomics signature. A time-independent validation cohort (n=48) was extracted from the 192 patients to evaluate the DL model with area under the receiver operating characteristic curve (AUC), specificity, and sensitivity. Meanwhile, a traditional radiomics model was constructed using least absolute shrinkage and selection operator (lasso) as comparisons with the DL-based model.

RESULTS : According to RECIST criteria, 131 patients were identified as responders with complete response, partial response and stable disease, while 61 patients were non-responders with progression disease. The selected predictive clinical factor turned out to be the carcinoembryonic antigen (CEA) level with AUC of 0.489 (95% confidence interval [CI], 0.380-0.599) and 0.558 (95% CI, 0.374-0.741) in the training and validation cohorts, respectively. The DL-based model provided better performance than the traditional classifier-based radiomics model with significantly higher AUC (training: 0.903 [95% CI, 0.851-0.955] vs.0.745 [95% CI, 0.659-0.831]; validation: 0.820 [95% CI, 0.681-0.959] vs. 0.598 [95% CI, 0.422-0.774]). The combination of DL-based model with the CEA level provided slightly increased performance with AUC of 0.935 [95% CI, 0.897-0.973] in the training cohort and 0.830 [95% CI, 0.688-0.973] in the validation cohort.

CONCLUSIONS : The developed DL-based radiomics model could improve the efficiency to predict the response to chemotherapy in CRLM, which may assist in subsequent personalized treatment decision-making in CRLM management.

Wei Jingwei, Cheng Jin, Gu Dongsheng, Chai Fan, Hong Nan, Wang Yi, Tian Jie


Chemotherapy, Colorectal Liver Metastases, Contrast-Enhanced Multidetector Computed Tomography, Deep Learning, Radiomics

General General

Introducing a hybrid artificial intelligence method for high-throughput modeling and optimizing plant tissue culture processes: the establishment of a new embryogenesis medium for chrysanthemum, as a case study.

In Applied microbiology and biotechnology

Data-driven models in a combination of optimization algorithms could be beneficial methods for predicting and optimizing in vitro culture processes. This study was aimed at modeling and optimizing a new embryogenesis medium for chrysanthemum. Three individual data-driven models, including multi-layer perceptron (MLP), adaptive neuro-fuzzy inference system (ANFIS), and support vector regression (SVR), were developed for callogenesis rate (CR), embryogenesis rate (ER), and somatic embryo number (SEN). Consequently, the best obtained results were used in the fusion process by a bagging method. For medium reformulation, effects of eight ionic macronutrients on CR, ER, and SEN and effects of four vitamins on SEN were evaluated using data fusion (DF)-non-dominated sorting genetic algorithm-II (NSGA-II) and DF-genetic algorithm (GA), respectively. Results showed that DF models with the highest R2 had superb performance in comparison with all other individual models. According to DF-NSGAII, the highest ER and SEN can be obtained from the medium containing 14.27 mM NH4+, 38.92 mM NO3-, 22.79 mM K+, 5.08 mM Cl-, 3.34 mM Ca2+, 1.67 mM Mg2+, 2.17 mM SO42-, and 1.44 mM H2PO4-. Based on the DF-GA model, the maximum SEN can be obtained from a medium containing 0.61 μM thiamine, 5.93 μM nicotinic acid, 0.25 μM biotin, and 0.26 μM riboflavin. The efficiency of the established-optimized medium was experimentally compared to Murashige and Skoog medium (MS) for embryogenesis of five chrysanthemum cultivars, and results indicated the efficiency of optimized medium over MS medium.Key points• MLP, SVR, and ANFIS were fused by a bagging method to develop a data fusion model.• NSGA-II and GA were linked to the data fusion model for establishing and optimizing a new embryogenesis medium.• The new culture medium (HNT) had better efficiency than MS medium.

Hesami Mohsen, Naderi Roohangiz, Tohidfar Masoud


Data fusion, Data-driven model, In vitro culture, Optimization algorithm, Sensitivity analysis

General General

Floral Complexity Traits as Predictors of Plant-Bee Interactions in a Mediterranean Pollination Web.

In Plants (Basel, Switzerland)

Despite intensive research, predicting pairwise species associations in pollination networks remains a challenge. The morphological fit between flowers and pollinators acts as a filter that allows only some species within the network to interact. Previous studies emphasized the depth of floral tubes as a key shape trait that explains the composition of their animal visitors. Yet, additional shape-related parameters, related to the handling difficulty of flowers, may be important as well. We analyzed a dataset of 2288 visits by six bee genera to 53 flowering species in a Mediterranean plant community. We characterized the plant species by five discrete shape parameters, which potentially affect their accessibility to insects: floral shape class, tube depth, symmetry, corolla segmentation and type of reproductive unit. We then trained a random forest machine-learning model to predict visitor identities, based on the shape traits. The model's predictor variables also included the Julian date on which each bee visit was observed and the year of observation, as proxies for within- and between-season variation in flower and bee abundance. The model attained a classification accuracy of 0.86 (AUC = 0.96). Using only shape parameters as predictors reduced its classification accuracy to 0.76 (AUC = 0.86), while using only the date and year variables resulted in a prediction accuracy of 0.69 (AUC = 0.80). Among the shape-related variables considered, flower shape class was the most important predictor of visitor identity in a logistic regression model. Our study demonstrates the power of machine-learning algorithms for understanding pollination interactions in a species-rich plant community, based on multiple features of flower morphology.

Ornai Alon, Keasar Tamar


complexity, flower morphology, machine learning, phenology, pollination network