MOTIVATION: Data scarcity limits the characterization of protein fitness landscapes and the development of accurate variant effect prediction models. To address this challenge, we introduce fitness translocation, a data augmentation strategy that generates synthetic variants for a target protein by leveraging variant fitness data previously measured in homologous proteins. Using embeddings from protein language models, the method computes the difference between each homolog variant and its wild type and applies these offsets to the target wild-type embedding to create synthetic variants in embedding space. RESULTS: We illustrate the utility of fitness translocation in the context of variant effect prediction on three protein families: IGPS, GFP, and SARS-CoV-2 spike proteins, across different models and training data sizes. Fitness translocation consistently improves predictive performance, particularly under limited training data, and is effective even when augmenting with remote homologs sharing as little as 35% sequence identity. These results illustrate how biologically grounded data augmentation can expand and diversify protein fitness landscapes, supporting more data-efficient protein engineering. AVAILABILITY AND IMPLEMENTATION: The code and datasets are available at https://github.com/adrienmialland/ProtFitTrans.

Machine learning-based protein mutational effect prediction is widely used in protein engineering and pathogenicity prediction, but training data scarcity remains a major challenge due to high costs of experimental measurements. A previous study proposed data augmentation using computational estimates by molecular simulation. However, this approach has been limited to predicting mutational effects on thermostability. Here, we present a new data augmentation method that combines molecular simulation with zero-shot prediction computed by protein language models. These computational estimates serve as 'weak' training data to supplement experimental training data. Our method dynamically adjusts the weight and inclusion of weak training data based on available experimental training data. This reduces potential negative impacts of weak training data while extending applicability to diverse protein properties such as binding affinity and enzymatic activity. Benchmark tests demonstrate that our method improves prediction accuracy particularly when experimental training data are scarce. These results indicate the capability of our approach to advance protein engineering and pathogenicity prediction in small data regimes.

Effective cancer subtype classification from multi-omics data remains challenging due to incomplete omics data and limited sample sizes. While graph convolutional networks (GCNs) have been used to incorporate inter-sample relationships for enhancing small-sample classification, their performance deteriorates when a certain omics modality is entirely missing. Here, we propose MOGEDN, a novel framework for cancer subtype classification using multi-omics encoder-decoder networks designed to reconstruct the latent features of missing omics data. The reconstructed features are integrated with available omics features to enable robust prediction under small-sample and missing-omics settings. We develop a step-wise algorithm to pretrain our model with diverse cancer types then to finetune for a specific cancer type while incorporating inter-sample and cross-omics dependencies. Evaluated on TCGA cancer datasets including subtypes with fewer than 50 samples, MOGEDEN consistently outperforms state-of-the-art baselines in accuracy and F1 scores. Moreover, MOGEDN's feature analysis provides two complementary biomarker sets: biomarkers shared across diverse cancer types in the pretraining phase; and biomarkers for a specific cancer type in the finetuning phase, facilitating model interpretability, and biological findings. These results highlight decoder-based imputation as a powerful approach to enhance multi-omics learning, delivering accurate classification, robust few-shot performance, and multi-scale biomarker discovery in incomplete multi-omics cohorts.

Accurate variant effect prediction has broad impacts on protein engineering. Recent machine learning approaches toward this end are based on representation learning, by which feature vectors are learned and generated from unlabeled sequences. However, it is unclear how to effectively learn evolutionary properties of an engineering target protein from homologous sequences, taking into account the protein's sequence-level structure called domain architecture (DA). Additionally, no optimal protocols are established for incorporating such properties into Transformer, the neural network well-known to perform the best in natural language processing research. This article proposes DA-aware evolutionary fine-tuning, or 'evotuning', protocols for Transformer-based variant effect prediction, considering various combinations of homology search, fine-tuning and sequence vectorization strategies. We exhaustively evaluated our protocols on diverse proteins with different functions and DAs. The results indicated that our protocols achieved significantly better performances than previous DA-unaware ones. The visualizations of attention maps suggested that the structural information was incorporated by evotuning without direct supervision, possibly leading to better prediction accuracy.

When this paper originally published, it was erroneously published alongside supplementary data from another paper. This has now been corrected online, and the correct supplementary data has been published. The publisher apologises for this error.

BACKGROUND: Previous studies have suggested deep learning to be a highly effective approach for screening lead compounds for new drugs. Several deep learning models have been developed by addressing the use of various kinds of fingerprints and graph convolution architectures. However, these methods are either advantageous or disadvantageous depending on whether they (1) can distinguish structural differences including chirality of compounds, and (2) can automatically discover effective features. RESULTS: We developed another deep learning model for compound classification. In this method, we constructed a distributed representation of compounds based on the SMILES notation, which linearly represents a compound structure, and applied the SMILES-based representation to a convolutional neural network (CNN). The use of SMILES allows us to process all types of compounds while incorporating a broad range of structure information, and representation learning by CNN automatically acquires a low-dimensional representation of input features. In a benchmark experiment using the TOX 21 dataset, our method outperformed conventional fingerprint methods, and performed comparably against the winning model of the TOX 21 Challenge. Multivariate analysis confirmed that the chemical space consisting of the features learned by SMILES-based representation learning adequately expressed a richer feature space that enabled the accurate discrimination of compounds. Using motif detection with the learned filters, not only important known structures (motifs) such as protein-binding sites but also structures of unknown functional groups were detected. CONCLUSIONS: The source code of our SMILES-based convolutional neural network software in the deep learning framework Chainer is available at http://www.dna.bio.keio.ac.jp/smiles/ , and the dataset used for performance evaluation in this work is available at the same URL.

This preliminary research investigates the effects of self-regulated learning support using the system "Self-regulator (SR)," and relationships between self-regulated learning awareness, learning behaviors, and perceived effects of SR. The results showed that the course with SR promoted "meet the deadline" awareness. The results of Spearman's correlation analysis revealed that procrastination awareness for high performance is one of the key factors for time management, which is an important factor of self-regulated learning.

BACKGROUND: Prediction of biochemical (metabolic) pathways has a wide range of applications, including the optimization of drug candidates, and the elucidation of toxicity mechanisms. Recently, several methods have been developed for pathway prediction to derive a goal compound from a start compound. However, these methods require high computational costs, and cannot perform comprehensive prediction of novel metabolic pathways. Our aim of this study is to develop a de novo prediction method for reconstructions of metabolic pathways and predictions of unknown biosynthetic pathways in the sense that it does not require any initial network such as KEGG metabolic network to be explored. RESULTS: We formulated pathway prediction between a start compound and a goal compound as the shortest path search problem in terms of the number of enzyme reactions applied. We propose an efficient search method based on A* algorithm and heuristic techniques utilizing Linear Programming (LP) solution for estimation of the distance to the goal. First, a chemical compound is represented by a feature vector which counts frequencies of substructure occurrences in the structural formula. Second, an enzyme reaction is represented as an operator vector by detecting the structural changes to compounds before and after the reaction. By defining compound vectors as nodes and operator vectors as edges, prediction of the reaction pathway is reduced to the shortest path search problem in the vector space. In experiments on the DDT degradation pathway, we verify that the shortest paths predicted by our method are biologically correct pathways registered in the KEGG database. The results also demonstrate that the LP heuristics can achieve significant reduction in computation time. Furthermore, we apply our method to a secondary metabolite pathway of plant origin, and successfully find a novel biochemical pathway which cannot be predicted by the existing method. For the reconstruction of a known biochemical pathway, our method is over 40 times as fast as the existing method. CONCLUSIONS: Our method enables fast and accurate de novo pathway predictions and novel pathway detection.

BACKGROUND: Clustering of unannotated transcripts is an important task to identify novel families of noncoding RNAs (ncRNAs). Several hierarchical clustering methods have been developed using similarity measures based on the scores of structural alignment. However, the high computational cost of exact structural alignment requires these methods to employ approximate algorithms. Such heuristics degrade the quality of clustering results, especially when the similarity among family members is not detectable at the primary sequence level. RESULTS: We describe a new similarity measure for the hierarchical clustering of ncRNAs. The idea is that the reliability of approximate algorithms can be improved by utilizing the information of suboptimal solutions in their dynamic programming frameworks. We approximate structural alignment in a more simplified manner than the existing methods. Instead, our method utilizes all possible sequence alignments and all possible secondary structures, whereas the existing methods only use one optimal sequence alignment and one optimal secondary structure. We demonstrate that this strategy can achieve the best balance between the computational cost and the quality of the clustering. In particular, our method can keep its high performance even when the sequence identity of family members is less than 60%. CONCLUSIONS: Our method enables fast and accurate clustering of ncRNAs. The software is available for download at http://bpla-kernel.dna.bio.keio.ac.jp/clustering/.

BACKGROUND: Computational prediction of noncoding RNAs (ncRNAs) is an important task in the post-genomic era. One common approach is to utilize the profile information contained in alignment data rather than single sequences. However, this strategy involves the possibility that the quality of input alignments can influence the performance of prediction methods. Therefore, the evaluation of the robustness against alignment errors is necessary as well as the development of accurate prediction methods. RESULTS: We describe a new method, called Profile BPLA kernel, which predicts ncRNAs from alignment data in combination with support vector machines (SVMs). Profile BPLA kernel is an extension of base-pairing profile local alignment (BPLA) kernel which we previously developed for the prediction from single sequences. By utilizing the profile information of alignment data, the proposed kernel can achieve better accuracy than the original BPLA kernel. We show that Profile BPLA kernel outperforms the existing prediction methods which also utilize the profile information using the high-quality structural alignment dataset. In addition to these standard benchmark tests, we extensively evaluate the robustness of Profile BPLA kernel against errors in input alignments. We consider two different types of error: first, that all sequences in an alignment are actually ncRNAs but are aligned ignoring their secondary structures; second, that an alignment contains unrelated sequences which are not ncRNAs but still aligned. In both cases, the effects on the performance of Profile BPLA kernel are surprisingly small. Especially for the latter case, we demonstrate that Profile BPLA kernel is more robust compared to the existing prediction methods. CONCLUSIONS: Profile BPLA kernel provides a promising way for identifying ncRNAs from alignment data. It is more accurate than the existing prediction methods, and can keep its performance under the practical situations in which the quality of input alignments is not necessarily high.

This paper proposes a multiple-loop array antenna (MLAA) for short-range pulse radars in the quasi-millimeter wave frequency band. The MLAA, which consists of multiple loop elements and two detour elements, has a tilt-beam characteristic, and is small in planar size. There are two schemes included in our method of applying the MLAA to a short-range pulse radar. One is a scheme that switches the beam direction by selecting the MLAA's feed points. The other is a scheme that switches between the two MLAAs with the fixed tilt-beam characteristic, the beam directions of which are different to each other. The advantage of using the switched-beam characteristic is that the number of sensors can be reduced. In this paper, we demonstrate the MLAA with the switched beam. The MLAA with a switch circuit using PIN diodes is measured to have the characteristics of 10 dBi in gain at 26 GHz, together with an effective switched-beam characteristic.

This paper presents a new coding method for 3-D geometry data using surface simplification and the wavelet transform. A polygonal mesh model is constructed from the connectivity information and the geometric data. We have previously presented a coding method for the structured geometry data structured on a 2-D lattice plane. In this paper we present a coding method for the structured geometry data including the vertices which were contracted by the surface simplification as extended nodes on the 2-D plane. We apply the shape-adaptive wavelet transform to the structured geometry data containing the extended nodes to obtain the wavelet coefficients. Defining the parent-children dependency of the wavelet coefficients among the different frequency bands in consideration of the extended nodes, we obtain the coded data by using SFQ. Some experiments for a statue showed that the proposed method gives good coding performance compared to TAGC scheme.

A multi-antenna system without any switching circuits, which is suitable for the CDMA handy phone is proposed. The proposed antenna system is compiled of the two quarter wavelength monopole antennas (QMPA) located at the top and bottom of the handy phone, and the two QMPAs are simultaneously fed through a power divider in phase and same amplitude. The radiation efficiency is measured in three conditions that upper, center and lower portions of the handy phone are held by user's hand in the talk position. The radiation efficiency is generally high in the all conditions even if one of the QMPAs is wrapped by the hand. The proposed antenna system is effective to reduce the influence by user' hand. Moreover, the phase difference in feed circuit is varied by inserting a phase sifter between the power divider and the QMPA. Although the radiation pattern in free space is changed by the phase difference, the improvement of radiation efficiency in the talk position can be obtained as in phase.

In this paper, we propose a circular polarized antenna, which is one-wavelength L-shaped structure with one feed point. The proposed antenna is composed of two elements, which have slightly different length, and the feed point is arranged on the corner point. The length of the two elements is adjusted and the circular polarization characteristics are calculated by using the moment method. As a result, we make clear the principle of the proposed antenna that the circular polarization can be obtained by adjusting design parameters so that the phase difference of excitation currents on the two elements is in agreement with the included angle between the two elements. Moreover, we make an experiment on a fabricated slot type model of the proposed antenna and demonstrated its favorable impedance matching characteristic. The L-shaped and simple feeding structure of the proposed antenna is suited to be built in the corner of front panel of small radio equipment for wireless LAN systems.

In this paper, a planar antenna and the 6-sector arrangement for a small card type mobile terminal used in the wireless local area network (WLAN) system are proposed. The radiation characteristics, particularly the tilted main beam in vertical plane and current distribution are calculated and discussed. We show that the 6-sector beams of half power width of 60 degrees are obtained by the proposed arrangement that is in a rectangular configuration of 10.6 mm /spl times/ 44 mm at 25 GHz band, and this is suited to a built-in antenna mounted on the edge of the card type mobile terminal like a PCMCIA card.