Multi-Task Learning of Object States and State-Modifying Actions from Web Videos.

We aim to learn to temporally localize object state changes and the corresponding state-modifying actions by observing people interacting with objects in long uncurated web videos. We introduce three principal contributions. First, we develop a self-supervised model for jointly learning state-modifying actions together with the corresponding object states from an uncurated set of videos from the Internet. The model is self-supervised by the causal ordering signal, i.e., initial object state manipulating action end state. Second, we explore alternative multi-task network architectures and identify a model that enables efficient joint learning of multiple object states and actions, such as pouring water and pouring coffee, together. Third, we collect a new dataset, named ChangeIt, with more than 2600 hours of video and 34 thousand changes of object states. We report results on an existing instructional video dataset COIN as well as our new large-scale ChangeIt dataset containing tens of thousands of long uncurated web videos depicting various interactions such as hole drilling, cream whisking, or paper plane folding. We show that our multi-task model achieves a relative improvement of 40% over the prior methods and significantly outperforms both image-based and video-based zero-shot models for this problem.

Mechanism of mitoribosomal small subunit biogenesis and preinitiation.

Mitoribosomes are essential for the synthesis and maintenance of bioenergetic proteins. Here we use cryo-electron microscopy to determine a series of the small mitoribosomal subunit (SSU) intermediates in complex with auxiliary factors, revealing a sequential assembly mechanism. The methyltransferase TFB1M binds to partially unfolded rRNA h45 that is promoted by RBFA, while the mRNA channel is blocked. This enables binding of METTL15 that promotes further rRNA maturation and a large conformational change of RBFA. The new conformation allows initiation factor mtIF3 to already occupy the subunit interface during the assembly. Finally, the mitochondria-specific ribosomal protein mS37 (ref. 1) outcompetes RBFA to complete the assembly with the SSU-mS37-mtIF3 complex2 that proceeds towards mtIF2 binding and translation initiation. Our results explain how the action of step-specific factors modulate the dynamic assembly of the SSU, and adaptation of a unique protein, mS37, links the assembly to initiation to establish the catalytic human mitoribosome.

Mitochondrial rRNA Methylation by Mettl15 Contributes to the Exercise and Learning Capability in Mice.

Mitochondrial translation is a unique relic of the symbiotic origin of the organelle. Alterations of its components cause a number of severe human diseases. Hereby we report a study of mice devoid of Mettl15 mitochondrial 12S rRNA methyltransferase, responsible for the formation of m4C839 residue (human numbering). Homozygous Mettl15-/- mice appeared to be viable in contrast to other mitochondrial rRNA methyltransferase knockouts reported earlier. The phenotype of Mettl15-/- mice is much milder than that of other mutants of mitochondrial translation apparatus. In agreement with the results obtained earlier for cell cultures with an inactivated Mettl15 gene, we observed accumulation of the RbfA factor, normally associated with the precursor of the 28S subunit, in the 55S mitochondrial ribosome fraction of knockout mice. A lack of Mettl15 leads to a lower blood glucose level after physical exercise relative to that of the wild-type mice. Mettl15-/- mice demonstrated suboptimal muscle performance and lower levels of Cox3 protein synthesized by mitoribosomes in the oxidative soleus muscles. Additionally, we detected decreased learning capabilities in the Mettl15-/- knockout mice in the tests with both positive and negative reinforcement. Such properties make Mettl15-/- knockout mice a suitable model for mild mitochondriopathies.

Learning to Answer Visual Questions from Web Videos.

Recent methods for visual question answering rely on large-scale annotated datasets. Manual annotation of questions and answers for videos, however, is tedious, expensive and prevents scalability. In this work, we propose to avoid manual annotation and generate a large-scale training dataset for video question answering making use of automatic cross-modal supervision. We leverage a question generation transformer trained on text data and use it to generate question-answer pairs from transcribed video narrations. Given narrated videos, we then automatically generate the HowToVQA69M dataset with 69M video-question-answer triplets. To handle the open vocabulary of diverse answers in this dataset, we propose a training procedure based on a contrastive loss between a video-question multi-modal transformer and an answer transformer. We introduce the zero-shot VideoQA task and the VideoQA feature probe evaluation setting and show excellent results. Furthermore, our method achieves competitive results on MSRVTT-QA, ActivityNet-QA, MSVD-QA and How2QA datasets. We also show that our approach generalizes to another source of web video and text data. We generate the WebVidVQA3M dataset from videos with alt-text annotations, and show its benefits for training VideoQA models. Finally, for a detailed evaluation we introduce iVQA, a new VideoQA dataset with reduced language bias and high-quality manual annotations.

Epitranscriptomics of Mammalian Mitochondrial Ribosomal RNA.

Modified nucleotides are present in all ribosomal RNA molecules. Mitochondrial ribosomes are unique to have a set of methylated residues that includes universally conserved ones, those that could be found either in bacterial or in archaeal/eukaryotic cytosolic ribosomes and those that are present exclusively in mitochondria. A single pseudouridine within the mt-rRNA is located in the peptidyltransferase center at a position similar to that in bacteria. After recent completion of the list of enzymes responsible for the modification of mammalian mitochondrial rRNA it became possible to summarize an evolutionary history, functional role of mt-rRNA modification enzymes and an interplay of the mt-rRNA modification and mitoribosome assembly process, which is a goal of this review.

METTL15 interacts with the assembly intermediate of murine mitochondrial small ribosomal subunit to form m4C840 12S rRNA residue.

Mammalian mitochondrial ribosomes contain a set of modified nucleotides, which is distinct from that of the cytosolic ribosomes. Nucleotide m4C840 of the murine mitochondrial 12S rRNA is equivalent to the dimethylated m4Cm1402 residue of Escherichia coli 16S rRNA. Here we demonstrate that mouse METTL15 protein is responsible for the formation of m4C residue of the 12S rRNA. Inactivation of Mettl15 gene in murine cell line perturbs the composition of mitochondrial protein biosynthesis machinery. Identification of METTL15 interaction partners revealed that the likely substrate for this RNA methyltransferase is an assembly intermediate of the mitochondrial small ribosomal subunit containing an assembly factor RBFA.

Mouse Trmt2B protein is a dual specific mitochondrial metyltransferase responsible for m5U formation in both tRNA and rRNA.

RNA molecules of all species contain modified nucleotides and particularly m5U residues. The vertebrate mitochondrial small subunit rRNA contains m5U nucleotide in a unique site. In this work we found an enzyme, TRMT2B, responsible for the formation of this nucleotide and m5U residues in a number of mitochondrial tRNA species. Inactivation of the Trmt2B gene leads to a reduction of the activity of respiratory chain complexes I, III and IV, containing the subunits synthesized by the mitochondrial protein synthesis apparatus. Comparative sequence analysis of m5U-specific RNA methyltransferases revealed an unusual evolutionary pathway of TRMT2B formation which includes consecutive substrate specificity switches from the large subunit rRNA to tRNA and then to the small subunit rRNA.

Long-Term Temporal Convolutions for Action Recognition.

Typical human actions last several seconds and exhibit characteristic spatio-temporal structure. Recent methods attempt to capture this structure and learn action representations with convolutional neural networks. Such representations, however, are typically learned at the level of a few video frames failing to model actions at their full temporal extent. In this work we learn video representations using neural networks with long-term temporal convolutions (LTC). We demonstrate that LTC-CNN models with increased temporal extents improve the accuracy of action recognition. We also study the impact of different low-level representations, such as raw values of video pixels and optical flow vector fields and demonstrate the importance of high-quality optical flow estimation for learning accurate action models. We report state-of-the-art results on two challenging benchmarks for human action recognition UCF101 (92.7%) and HMDB51 (67.2%).

Learning from Narrated Instruction Videos.

Automatic assistants could guide a person or a robot in performing new tasks, such as changing a car tire or repotting a plant. Creating such assistants, however, is non-trivial and requires understanding of visual and verbal content of a video. Towards this goal, we here address the problem of automatically learning the main steps of a task from a set of narrated instruction videos. We develop a new unsupervised learning approach that takes advantage of the complementary nature of the input video and the associated narration. The method sequentially clusters textual and visual representations of a task, where the two clustering problems are linked by joint constraints to obtain a single coherent sequence of steps in both modalities. To evaluate our method, we collect and annotate a new challenging dataset of real-world instruction videos from the Internet. The dataset contains videos for five different tasks with complex interactions between people and objects, captured in a variety of indoor and outdoor settings. We experimentally demonstrate that the proposed method can automatically discover, learn and localize the main steps of a task in input videos.

Co-expression of glucose-6-phosphate dehydrogenase and acyl-CoA binding protein enhances lipid accumulation in the yeast Yarrowia lipolytica.

The oleaginous yeast Yarrowia lipolytica is a convenient model for investigating lipid biosynthesis and for engineering high lipid accumulated strains. In this organism, the pentose phosphate pathway is the major source of NADPH for lipid biosynthesis. Thus, we over-expressed gene encoding NADP+-dependent glucose-6-phosphate dehydrogenase (ZWF1) in a strain deficient in peroxisome biogenesis. However, this strategy suppressed growth during cultivation under lipogenic conditions and did not significantly increase lipid accumulation. Remarkably, co-expression of gene encoding acyl-CoA binding protein (ACBP), which functions as an intracellular acyl-CoA transporter and acyl-CoA-pool former, restored growth. Co-expression of ZWF1 and ACBP increased the lipid content to 30% of dry cell weight via de novo lipid synthesis. In comparison to wild type, the engineered strain accumulated 41% more lipids with a higher ratio of saturated to unsaturated fatty acids.

Metabolic evolution and (13) C flux analysis of a succinate dehydrogenase deficient strain of Yarrowia lipolytica.

Bio-based succinic acid production can redirect industrial chemistry processes from using limited hydrocarbons to renewable carbohydrates. A fermentation process that does not require pH-titrating agents will be advantageous to the industry. Previously, a Yarrowia lipolytica strain that was defective for succinate dehydrogenase was constructed and was found to accumulate up to 17.5 g L(-1) of succinic acid when grown on glycerol without buffering. Here, a derivative mutant was isolated that produced 40.5 g L(-1) of succinic acid in 36 h with a yield of 0.32 g g(-1) glycerol. A combination approach of induced mutagenesis and metabolic evolution allowed isolation of another derivative that could utilize glucose efficiently and accumulated 50.2 g L(-1) succinic acid in 54 h with a yield of 0.43 g g(-1) . The parent strain of these isolated mutants was used for [1,6-(13) C2 ]glucose assimilation analysis. At least 35% glucose was estimated to be utilized through the pentose phosphate pathway, while ≥84% succinic acid was formed through the oxidative branch of the tricarboxylic acid cycle. Biotechnol. Bioeng. 2016;113: 2425-2432. © 2016 Wiley Periodicals, Inc.

N6-Methylated Adenosine in RNA: From Bacteria to Humans.

N6-methyladenosine (m(6)A) is ubiquitously present in the RNA of living organisms from Escherichia coli to humans. Methyltransferases that catalyze adenosine methylation are drastically different in specificity from modification of single residues in bacterial ribosomal or transfer RNA to modification of thousands of residues spread among eukaryotic mRNA. Interactions that are formed by m(6)A residues range from RNA-RNA tertiary contacts to RNA-protein recognition. Consequences of the modification loss might vary from nearly negligible to complete reprogramming of regulatory pathways and lethality. In this review, we summarized current knowledge on enzymes that introduce m(6)A modification, ways to detect m(6)A presence in RNA and the functional role of this modification everywhere it is present, from bacteria to humans.

Pose Estimation and Segmentation of Multiple People in Stereoscopic Movies.

We describe a method to obtain a pixel-wise segmentation and pose estimation of multiple people in stereoscopic videos. This task involves challenges such as dealing with unconstrained stereoscopic video, non-stationary cameras, and complex indoor and outdoor dynamic scenes with multiple people. We cast the problem as a discrete labelling task involving multiple person labels, devise a suitable cost function, and optimize it efficiently. The contributions of our work are two-fold: First, we develop a segmentation model incorporating person detections and learnt articulated pose segmentation masks, as well as colour, motion, and stereo disparity cues. The model also explicitly represents depth ordering and occlusion. Second, we introduce a stereoscopic dataset with frames extracted from feature-length movies "StreetDance 3D" and "Pina". The dataset contains 587 annotated human poses, 1,158 bounding box annotations and 686 pixel-wise segmentations of people. The dataset is composed of indoor and outdoor scenes depicting multiple people with frequent occlusions. We demonstrate results on our new challenging dataset, as well as on the H2view dataset from (Sheasby et al. ACCV 2012).

Sustainable production of biologically active molecules of marine based origin.

The marine environment offers both economic and scientific potential which are relatively untapped from a biotechnological point of view. These environments whilst harsh are ironically fragile and dependent on a harmonious life form balance. Exploitation of natural resources by exhaustive wild harvesting has obvious negative environmental consequences. From a European industry perspective marine organisms are a largely underutilised resource. This is not due to lack of interest but due to a lack of choice the industry faces for cost competitive, sustainable and environmentally conscientious product alternatives. Knowledge of the biotechnological potential of marine organisms together with the development of sustainable systems for their cultivation, processing and utilisation are essential. In 2010, the European Commission recognised this need and funded a collaborative RTD/SME project under the Framework 7-Knowledge Based Bio-Economy (KBBE) Theme 2 Programme 'Sustainable culture of marine microorganisms, algae and/or invertebrates for high value added products'. The scope of that project entitled 'Sustainable Production of Biologically Active Molecules of Marine Based Origin' (BAMMBO) is outlined. Although the Union is a global leader in many technologies, it faces increasing competition from traditional rivals and emerging economies alike and must therefore improve its innovation performance. For this reason innovation is placed at the heart of a European Horizon 2020 Strategy wherein the challenge is to connect economic performance to eco performance. This article provides a synopsis of the research activities of the BAMMBO project as they fit within the wider scope of sustainable environmentally conscientious marine resource exploitation for high-value biomolecules.

Production of recombinant Rhizopus oryzae lipase by the yeast Yarrowia lipolytica results in increased enzymatic thermostability.

The gene encoding Rhizopus oryzae lipase (ROL) was expressed in the non-conventional yeast Yarrowia lipolytica under the control of the strong inducible XPR2 gene promoter. The effects of three different preprosequence variants were examined: a preprosequence of the Y. lipolytica alkaline extracellular protease (AEP) encoded by XPR2, the native preprosequence of ROL, and a hybrid variant of the presequence of AEP and the prosequence of ROL. Lipase production was highest (7.6 U/mL) with the hybrid prepropeptide. The recombinant protein was purified by ion-exchange chromatography. The ROL included 28 amino acids of the C-terminal region of the prosequence, indicating that proteolytic cleavage occurred below the KR site through the activity of the Kex2-like endoprotease. The optimum temperature for recombinant lipase activity was between 30 and 40 °C, and the optimum pH was 7.5. The enzyme was shown not to be glycosylated. Furthermore, recombinant ROL exhibited greater thermostability than previously reported, with the enzyme retaining 64% of its hydrolytic activity after 30 min of incubation at 55 °C.

Is it possible to produce succinic acid at a low pH?

Bio-based succinate is still a matter of special emphasis in biotechnology and adjacent research areas. The vast majority of natural and engineered producers are bacterial strains that accumulate succinate under anaerobic conditions. Recently, we succeeded in obtaining an aerobic yeast strain capable of producing succinic acid at low pH. Herein, we discuss some difficulties and advantages of microbial pathways producing "succinic acid" rather than "succinate." It was concluded that the peculiar properties of the constructed yeast strain could be clarified in view of a distorted energy balance. There is evidence that in an acidic environment, the majority of the cellular energy available as ATP will be spent for proton and anion efflux. The decreased ATP:ADP ratio could essentially reduce the growth rate or even completely inhibit growth. In the same way, the preference of this elaborated strain for certain carbon sources could be explained in terms of energy balance. Nevertheless, the opportunity to exclude alkali and mineral acid waste from microbial succinate production seems environmentally friendly and cost-effective.

Efficient cell surface display of Lip2 lipase using C-domains of glycosylphosphatidylinositol-anchored cell wall proteins of Yarrowia lipolytica.

The cell surface display of enzymes is of great interest because of its simplified purification stage and the possibility for recycling in industrial processes. In this study, we have focused on the cell wall immobilization of Yarrowia lipolytica Lip2 protein--an enzyme that has a wide technological application. By genome analysis of Y. lipolytica in addition to already characterized Ylcwp1, we identified five putative open reading frames encoding glycosylphosphatidylinositol-anchored proteins. Lip2 translation fusion with the carboxyl termini of these proteins revealed that all proteins were capable of immobilizing lipase in active form on the cell surface. The highest level of cell-bound lipase activity was achieved using C-domains encoded by YlCWP1, YlCWP3 (YALI0D27214g) and YlCWP6 (YALI0F18282g) comprising 16,173 ± 1,800, 18,785 ± 1,130 and 17,700 ± 2,101 U/g dry cells, respectively. To the best of our knowledge, these results significantly exceed the highest cell-bound lipase activity previously reported for engineered Saccharomyces cerevisiae and Pichia pastoris strains. Furthermore, the lyophilized biomass retained the activity and was robust to collecting/resuspending procedures. Nevertheless, in most cases, a substantial amount of lipase activity was also found in the growth medium. Further work will be necessary to better understand the nature of this phenomenon.

View-independent action recognition from temporal self-similarities.

This paper addresses recognition of human actions under view changes. We explore self-similarities of action sequences over time and observe the striking stability of such measures across views. Building upon this key observation, we develop an action descriptor that captures the structure of temporal similarities and dissimilarities within an action sequence. Despite this temporal self-similarity descriptor not being strictly view-invariant, we provide intuition and experimental validation demonstrating its high stability under view changes. Self-similarity descriptors are also shown to be stable under performance variations within a class of actions when individual speed fluctuations are ignored. If required, such fluctuations between two different instances of the same action class can be explicitly recovered with dynamic time warping, as will be demonstrated, to achieve cross-view action synchronization. More central to the current work, temporal ordering of local self-similarity descriptors can simply be ignored within a bag-of-features type of approach. Sufficient action discrimination is still retained in this way to build a view-independent action recognition system. Interestingly, self-similarities computed from different image features possess similar properties and can be used in a complementary fashion. Our method is simple and requires neither structure recovery nor multiview correspondence estimation. Instead, it relies on weak geometric properties and combines them with machine learning for efficient cross-view action recognition. The method is validated on three public data sets. It has similar or superior performance compared to related methods and it performs well even in extreme conditions, such as when recognizing actions from top views while using side views only for training.

Production of succinic acid at low pH by a recombinant strain of the aerobic yeast Yarrowia lipolytica.

Biotechnological production of weak organic acids such as succinic acid is most economically advantageous when carried out at low pH. Among naturally occurring microorganisms, several bacterial strains are known to produce considerable amounts of succinic acid under anaerobic conditions but they are inefficient in performing the low-pH fermentation due to their physiological properties. We have proposed therefore a new strategy for construction of an aerobic eukaryotic producer on the basis of the yeast Yarrowia lipolytica with a deletion in the gene coding one of succinate dehydrogenase subunits. Firstly, an original in vitro mutagenesis-based approach was proposed to construct strains with Ts mutations in the Y. lipolytica SDH1 gene. These mutants were used to optimize the composition of the media for selection of transformants with the deletion in the Y. lipolytica SDH2 gene. Surprisingly, the defects of each succinate dehydrogenase subunit prevented the growth on glucose but the mutant strains grew on glycerol and produced succinate in the presence of the buffering agent CaCO(3). Subsequent selection of the strain with deleted SDH2 gene for increased viability allowed us to obtain a strain capable of accumulating succinate at the level of more than 45 g L(-1) in shaking flasks with buffering and more than 17 g L(-1) without buffering. The possible effect of the mutations on the utilization of different substrates and perspectives of constructing an industrial producer is discussed.