A Bayesian machine scientist to aid in the solution of challenging scientific problems.
Ontology highlight
ABSTRACT:
SUBMITTER: Guimera R
PROVIDER: S-EPMC6994216 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:Almost all machine learning (ML) is based on representing examples using intrinsic features. When there are multiple related ML problems (tasks), it is possible to transform these features into extrinsic features by first training ML models on other tasks and letting them each make predictions for each example of the new task, yielding a novel representation. We call this transformational ML (TML). TML is very closely related to, and synergistic with, transfer learning, multitask learning, and stacking. TML is applicable to improving any nonlinear ML method. We tested TML using the most important classes of nonlinear ML: random forests, gradient boosting machines, support vector machines, k-nearest neighbors, and neural networks. To ensure the generality and robustness of the evaluation, we utilized thousands of ML problems from three scientific domains: drug design, predicting gene expression, and ML algorithm selection. We found that TML significantly improved the predictive performance of all the ML methods in all the domains (4 to 50% average improvements) and that TML features generally outperformed intrinsic features. Use of TML also enhances scientific understanding through explainable ML. In drug design, we found that TML provided insight into drug target specificity, the relationships between drugs, and the relationships between target proteins. TML leads to an ecosystem-based approach to ML, where new tasks, examples, predictions, and so on synergistically interact to improve performance. To contribute to this ecosystem, all our data, code, and our ∼50,000 ML models have been fully annotated with metadata, linked, and openly published using Findability, Accessibility, Interoperability, and Reusability principles (∼100 Gbytes).
| S-EPMC8670494 | biostudies-literature
Project description:BackgroundUsing cDNA copies of transcripts and corresponding genomic sequences from the Berkeley Drosophila Genome Project, a set of 24,753 donor and acceptor splice sites were computed with a scanning algorithm that tested for single nucleotide insertion, deletion and substitution polymorphisms. Using this dataset, we developed a progressive partitioning approach to examining the effects of challenging the spliceosome system.ResultsOur analysis shows that information content increases near splice sites flanking progressively longer introns and exons, suggesting that longer splice elements require stronger binding of spliceosome components. Information also increases at splice sites near very short introns and exons, suggesting that short splice elements have crowding problems. We observe that the information found at individual splice sites depends upon a balance of splice element lengths in the vicinity, including both flanking and non-adjacent introns and exons.ConclusionThese results suggest an interdependence of multiple splicing events along the pre-mRNA, which may have implications for how the macromolecular spliceosome machine processes sets of neighboring splice sites.
| S-EPMC1431713 | biostudies-literature
Project description:The interpretation of archaeological features often requires a combined methodological approach in order to make the most of the material record, particularly from sites where this may be limited. In practice, this requires the consultation of different sources of information in order to cross validate findings and combat issues of ambiguity and equifinality. However, the application of a multiproxy approach often generates incompatible data, and might therefore still provide ambiguous results. This paper explores the potential of a simple digital framework to increase the explanatory power of multiproxy data by enabling the incorporation of incompatible, ambiguous datasets in a single model. In order to achieve this, Bayesian confirmation was used in combination with decision trees. The results of phytolith and geochemical analyses carried out on soil samples from ephemeral sites in Jordan are used here as a case study. The combination of the two datasets as part of a single model enabled us to refine the initial interpretation of the use of space at the archaeological sites by providing an alternative identification for certain activity areas. The potential applications of this model are much broader, as it can also help researchers in other domains reach an integrated interpretation of analysis results by combining different datasets.
| S-EPMC8011785 | biostudies-literature
Project description:This paper reviews some of the challenges posed by the huge growth of experimental data generated by the new generation of large-scale experiments at UK national facilities at the Rutherford Appleton Laboratory (RAL) site at Harwell near Oxford. Such 'Big Scientific Data' comes from the Diamond Light Source and Electron Microscopy Facilities, the ISIS Neutron and Muon Facility and the UK's Central Laser Facility. Increasingly, scientists are now required to use advanced machine learning and other AI technologies both to automate parts of the data pipeline and to help find new scientific discoveries in the analysis of their data. For commercially important applications, such as object recognition, natural language processing and automatic translation, deep learning has made dramatic breakthroughs. Google's DeepMind has now used the deep learning technology to develop their AlphaFold tool to make predictions for protein folding. Remarkably, it has been able to achieve some spectacular results for this specific scientific problem. Can deep learning be similarly transformative for other scientific problems? After a brief review of some initial applications of machine learning at the RAL, we focus on challenges and opportunities for AI in advancing materials science. Finally, we discuss the importance of developing some realistic machine learning benchmarks using Big Scientific Data coming from several different scientific domains. We conclude with some initial examples of our 'scientific machine learning' benchmark suite and of the research challenges these benchmarks will enable. This article is part of a discussion meeting issue 'Numerical algorithms for high-performance computational science'.
| S-EPMC7015290 | biostudies-literature
Project description:BackgroundScientific and professional development opportunities for early career scientists in low- and middle- income countries (LMICs) are limited and not consistent. There is a disproportionately low number of biomedical and clinical researchers in LMIC's relative to their high burden of disease, a disparity that is aggravated by emigration of up to 70% of scientists from their countries of birth for education and employment elsewhere. To help address this need, a novel University-accredited, immersive fellowship program was established by a large public-academic-private network. We sought to describe the program and summarize progress and lessons learned over its first 7-years.MethodsHallmarks of the program are a structured learning curriculum and bespoke research activities tailored to the needs of each fellow. Research projects expose the scientists to state-of-the-art methodologies and leading experts in their fields while also ensuring that learnings are implementable within their home infrastructure. Fellows run seminars on drug discovery and development that reinforce themes of scientific leadership and teamwork together with practical modules on addressing healthcare challenges within their local systems. Industry mentors achieve mutual learning to better understand healthcare needs in traditionally underserved settings. We evaluated the impact of the program through an online survey of participants and by assessing research output.ResultsMore than 140 scientists and clinicians from 25 countries participated over the 7-year period. Evaluation revealed strong evidence of knowledge and skills transfer, and beneficial self-reported impact on fellow's research output and career trajectories. Examples of program impact included completion of post-graduate qualifications; establishment and implementation of good laboratory- and clinical- practice mechanisms; and becoming lead investigators in local programs. There was a high retention of fellows in their home countries (>?75%) and an enduring professional network among the fellows and their mentors.ConclusionsOur experience demonstrates an example for how multi-sectoral partners can contribute to scientific and professional development of researchers in LMICs and supports the idea that capacity-building efforts should be tailored to the specific needs of beneficiaries to be maximally effective. Lessons learned may be applied to the design and conduct of other programs to strengthen science ecosystems in LMICs.
| S-EPMC6180641 | biostudies-literature
Project description:In this article, we focus on the potential influence of a scientist's advocacy position on the public's perceived credibility of scientists as a whole. Further, we examine how the scientist's solution position (information only, non-controversial, and controversial) affects the public's perception of the scientist's motivation for sharing information about specific issues (flu, marijuana, climate change, severe weather). Finally, we assess how perceived motivations mediate the relationship between solution position and credibility. Using data from a quota sample of American adults obtained by Qualtrics (n = 2,453), we found that in some conditions advocating for a solution positively predicted credibility, while in one condition, it negatively predicted scientist credibility. We also found that the influence of solution position on perceived credibility was mediated by several motivation perceptions; most notably through perception that the scientist was motivated to: (a) serve the public and (b) persuade the public. Further results and implications are discussed.
| S-EPMC5685625 | biostudies-literature
Project description:Boolean satisfiability is a propositional logic problem of interest in multiple fields, e.g., physics, mathematics, and computer science. Beyond a field of research, instances of the SAT problem, as it is known, require efficient solution methods in a variety of applications. It is the decision problem of determining whether a Boolean formula has a satisfying assignment, believed to require exponentially growing time for an algorithm to solve for the worst-case instances. Yet, the efficient solution of many classes of Boolean formulae eludes even the most successful algorithms, not only for the worst-case scenarios, but also for typical-case instances. Here, we introduce a memory-assisted physical system (a digital memcomputing machine) that, when its non-linear ordinary differential equations are integrated numerically, shows evidence for polynomially-bounded scalability while solving "hard" planted-solution instances of SAT, known to require exponential time to solve in the typical case for both complete and incomplete algorithms. Furthermore, we analytically demonstrate that the physical system can efficiently solve the SAT problem in continuous time, without the need to introduce chaos or an exponentially growing energy. The efficiency of the simulations is related to the collective dynamical properties of the original physical system that persist in the numerical integration to robustly guide the solution search even in the presence of numerical errors. We anticipate our results to broaden research directions in physics-inspired computing paradigms ranging from theory to application, from simulation to hardware implementation.
| S-EPMC7665037 | biostudies-literature
Project description:Backpropagation has been regarded as the most favorable algorithm for training artificial neural networks. However, it has been criticized for its biological implausibility because its learning mechanism contradicts the human brain. Although backpropagation has achieved super-human performance in various machine learning applications, it often shows limited performance in specific tasks. We collectively referred to such tasks as machine-challenging tasks (MCTs) and aimed to investigate methods to enhance machine learning for MCTs. Specifically, we start with a natural question: Can a learning mechanism that mimics the human brain lead to the improvement of MCT performances? We hypothesized that a learning mechanism replicating the human brain is effective for tasks where machine intelligence is difficult. Multiple experiments corresponding to specific types of MCTs where machine intelligence has room to improve performance were performed using predictive coding, a more biologically plausible learning algorithm than backpropagation. This study regarded incremental learning, long-tailed, and few-shot recognition as representative MCTs. With extensive experiments, we examined the effectiveness of predictive coding that robustly outperformed backpropagation-trained networks for the MCTs. We demonstrated that predictive coding-based incremental learning alleviates the effect of catastrophic forgetting. Next, predictive coding-based learning mitigates the classification bias in long-tailed recognition. Finally, we verified that the network trained with predictive coding could correctly predict corresponding targets with few samples. We analyzed the experimental result by drawing analogies between the properties of predictive coding networks and those of the human brain and discussing the potential of predictive coding networks in general machine learning.
| S-EPMC9709416 | biostudies-literature
Project description:[This corrects the article DOI: 10.1371/journal.pone.0187511.].
| S-EPMC6338381 | biostudies-literature
Project description:Clustering problems (including the clustering of individuals into outcrossing populations, hybrid generations, full-sib families and selfing lines) have recently received much attention in population genetics. In these clustering problems, the parameter of interest is a partition of the set of sampled individuals--the sample partition. In a fully Bayesian approach to clustering problems of this type, our knowledge about the sample partition is represented by a probability distribution on the space of possible sample partitions. As the number of possible partitions grows very rapidly with the sample size, we cannot visualize this probability distribution in its entirety, unless the sample is very small. As a solution to this visualization problem, we recommend using an agglomerative hierarchical clustering algorithm, which we call the exact linkage algorithm. This algorithm is a special case of the maximin clustering algorithm that we introduced previously. The exact linkage algorithm is now implemented in our software package PartitionView. The exact linkage algorithm takes the posterior co-assignment probabilities as input and yields as output a rooted binary tree, or more generally, a forest of such trees. Each node of this forest defines a set of individuals, and the node height is the posterior co-assignment probability of this set. This provides a useful visual representation of the uncertainty associated with the assignment of individuals to categories. It is also a useful starting point for a more detailed exploration of the posterior distribution in terms of the co-assignment probabilities.
| S-EPMC2705916 | biostudies-literature