Search

Background Identifying miRNA and disease associations helps us understand disease mechanisms of action from the molecular level. However, it is usually blind, time-consuming, and small-scale based on biological experiments. Hence, developing computational methods to predict unknown miRNA and disease associations is becoming increasingly important. Results In this work, we develop a computational framework called SMALF to predict unknown miRNA-disease associations. SMALF first utilizes a stacked autoencoder to learn miRNA latent feature and disease latent feature from the original miRNA-disease association matrix. Then, SMALF obtains the feature vector of representing miRNA-disease by integrating miRNA functional similarity, miRNA latent feature, disease semantic similarity, and disease latent feature. Finally, XGBoost is utilized to predict unknown miRNA-disease associations. We implement cross-validation experiments. Compared with other state-of-the-art methods, SAMLF achieved the best AUC value. We also construct three case studies, including hepatocellular carcinoma, colon cancer, and breast cancer. The results show that 10, 10, and 9 out of the top ten predicted miRNAs are verified in MNDR v3.0 or miRCancer, respectively. Conclusion The comprehensive experimental results demonstrate that SMALF is effective in identifying unknown miRNA-disease associations.

Arterio Venous Vascular Access (AVA) stenosis is a common complication in hemodialysis patients. Clinically, AVA stenosis happens when the cross-sectional area is reduced to less than 50% of the normal area. In order to highlight the correlation between the power spectral features of AVA Phonoangiography signals (PCG) and degree of stenosis (DOS), In-vitro BioPhysical Simulation Model (BPSM) is used to control individual conditions. Previous studies have pointed out that the features of PCG can be used to detect stenosis, but there were differences in the specific frequency band range. In this study, a method for extracting the power spectral features of PCG based on MUltiple SIgnal Classification power spectrum estimation algorithm (MUSIC) is proposed. This method has a high resolution for the high-frequency low-energy sound caused by stenosis. Using the proposed method, a strong correlation is found between the frequency peak near 820 Hz (820 ±70 Hz) and the AVA stenosis. Based on the above feature extraction method, a support vector machine (SVM) classification model is trained on data obtained on the BPSM. Finally, using MUSIC features extraction model and SVM classification model, the correct classification rate on BPSM data is 96.4%, and the SVM model is validated on 19 clinical measured data, the accuracy is 84.2%.

This study proposes a phonoangiography (PCG) signal processing method to help non-invasively diagnose Arteriovenous Graft (AVG) stenosis in patients with End Stage Renal Disease (ESRD). Pre-existing binary- classification studies for Degrees Of Stenosis (DOS) indicated that the presence of high frequency acoustic signals characterizes stenosis AVG, but the specific range they pointed out vary from each other, including 300Hz to 600Hz, 400Hz or more, 700Hz. In order to obtain the true relationship between the frequency characteristics of the acoustic signal and DOS of AVG, our team proposed a multi-position sequence feature (MPSF) extraction method, the judgement of DOS has also been expanded from binary-classification to multi-level classification. This method uses a physical model built by our team to simulate normal AVG and AVG with different DOS, we selected upstream 7cm and downstream 7cm of the stenosis section as two measurement positions to collect AVG acoustic signal. The components of the AVG acoustic signal below 200 Hz are filtered using singular spectrum analysis (SSA), and the remaining part is subjected to power spectrum estimation using the Welch method. The frequency spectrum energy obtained from the two measurement positions is combined into MPSF to be trained in the LSTM neural network, thereby realizing multi-level classification of different DOS. Tests on AVG physical models and clinical patient data sets show that, compared to previous studies, this method has the better effect on diagnosis of DOS.