Biomedical researchers often need to combine evidence from heterogeneous sources, including REST APIs, form-based deep web databases, semi-structured web pages, and downloadable files. Existing workflow systems support reproducible analysis but require manually specified pipelines, while LLM-based agents can issue tool calls but often lack explicit data dependencies, provenance, and controlled exploration. We present BioWeaver, an adaptive workflow orchestration system that converts natural-language biomedical questions into executable scientific workflow graphs. Each graph represents typed retrieval, refinement, and synthesis steps over a catalog of heterogeneous connectors. During execution, BioWeaver uses Progressive Data Refinement (PDR) to expand the workflow when intermediate results reveal useful follow-up records, while a composite information-gain heuristic and human-in-the-loop checkpoints control exploration depth. A unified connector abstraction supports APIs, browser-automated forms, HTML tables, and downloadable files under a common execution model. We evaluate BioWeaver on 45 queries across two biomedical question sets with gold-standard answers. Results show that BioWeaver improves overall answer quality by 6.2–14.5% over LLM baselines and by 41.4–53.8% over a ReAct agent. Ablation studies show that runtime refinement, planner knowledge, and information-gain stopping each contribute to system effectiveness. These results suggest that adaptive workflow graphs provide a practical foundation for reproducible, source-grounded biomedical data integration across heterogeneous and deep web resources.

In this paper we present a method for enhancing Question Answering (QA) systems by iteratively improving Knowledge Graphs (KGs) with a focus on maintaining monotonicity in the enhancement process. We introduce a mathematical framework employing functions τ and ϕ, where τ transforms text T into a KG K, and ϕ generates an answer from T for a given question. We propose that augmenting K with domain-specific information, denoted as ΔK, leads to a more accurate approximation of the expected answer, adhering to the principle that each enhancement either maintains or improves answer quality. This concept is formalized as ϕ− 1(ϕ(T) ∪ ΔK) yielding better results than ϕ− 1(ϕ(T)). The paper elaborates on this process with practical examples, demonstrating how KG enhancements, under the constraints of monotonicity, lead to successive improvements in the Question Answering (QA) system.

Designing novel materials and analyzing their properties is a computation intensive process. Increasingly modern machine learning techniques are being exploited in contemporary research to expedite and advance materials studies. One powerful tool available to researchers is the body of scientific knowledge that aids in selecting design models, algorithms, meta-data, and visualization tools to process and analyze experimental and empirical data for an iterative design process, potentially involving a human in the loop, In this preliminary research paper, our goal is to introduce a new machine learning platform, called MatFlow, for automated and knowledge driven design of novel materials and their usage. We outline its architecture and illustrate its functionality with an application in Transition Metal Dichalcogenide (TMD) Heterostructures design of electronic and energy devices.

The adoption and availability of diverse application design and support platforms are making generic scientific application orchestration increasingly difficult. In such an evolving environment, higher level abstractions of design primitives are critically important using which end users have a chance to craft their own applications without a complete technical grasp of the lower level details. In this research, we introduce a novel scientific workflow design platform that supports high level tools for data integration, process description and analytics based on a visual language for naive users and advanced options for computing savvy programmers in one single platform, called VisFlow. We describe its salient features and advantages using a complex scientific application in natural resources and ecology. Video: https://youtu.be/ 2YSYVyOuuk.

The abundance of predicted and mined but uncertain biological data show huge needs for massive, efficient and scalable curation efforts. The human expertise warranted by any successful curation enterprize is often economically prohibitive especially for speculative end user queries that may not ultimately bear fruit. So the challenge remains in devising a low cost engine capable of delivering fast but tentative annotation and curation of a set of data items that can be authoritatively validated by experts later demanding significantly small investment. The aim thus is to make a large volume of predicted data available for use as early as possible with an acceptable degree of confidence in their accuracy while the curation continues. In this paper, we present a novel approach to annotation and curation of biological database contents using crowd computing. The technical contribution is in the identification and management of trust of mechanical turks, and support for ad hoc declarative queries, both of which are leveraged to support reliable analytics using noisy predicted interactions.

The vast majority of phylogenetic databases do not support a declarative querying platform using which their contents can be flexibly and conveniently accessed. The template based query interfaces they support do not allow arbitrary speculative queries. While a small number of graph query languages such as XQuery, Cypher and GraphQL exist for computer savvy users, most are too general and complex to be useful for biologists, and too inefficient for large phylogeny querying. In this paper, we discuss a recently introduced visual query language, called PhyQL, that leverages phylogeny specific properties to support essential and powerful constructs for a large class of phylogentic queries. Its deductive reasoner based implementation offers opportunities for a wide range of pruning strategies to speed up processing using query specific optimization and thus making it suitable for large phylogeny querying. A hybrid optimization technique that exploits a set of indices and "graphlet" partitioning is discussed. A "fail soonest" strategy is used to avoid hopeless processing and is shown to produce dividends.

The deep web is very large and diverse and queries evaluated against the deep web can provide great value. While there have been attempts at accessing the data in the deep web, these are clever "one-of'' systems and techniques. In this paper, we describe an ongoing research of a generic structured query model that can be used against the deep web. Using this query model, the contributions of a community of researchers can be combined freely, leading to a system that can be improved incrementally each time someone develops a specific novel technique to improve a particular operator.

The cost of reachability query computation using traditional algorithms such as depth first search or transitive closure has been found to be prohibitive and unacceptable in massive graphs such as biological interaction networks, or pathways. Contemporary solutions mainly take two distinct approaches - precompute reachability in the form of transitive closure (trade space for time) or use state space search (trade time for space). A middle ground among the two approaches has recently gained popularity. It precomputes part of the reachability information as a complex index so that most queries can be answered within a reasonable time. In this approach, the main cost now is creation of the index, and response generation using it as well as the space needed to materialize the structure. Most contemporary solutions favor a combination of these costs to be efficient for a class of applications. In this paper, we propose a hierarchical index based on graph segmentation to reduce index size without sacrificing query efficiency. We present experimental evidence to show that our approach can achieve significant space savings, and improve efficiency. We also show that our index need not be rebuilt for a large class of updates, a feature missing in all other contemporary systems.

Ontologies, formal specifications of domains, have evolved in recent years as a leading tool in representing and interpreting Web data. The OntoBuilder project supports the extraction of ontologies from Web search interfaces, ranging from simple search engine forms to multiple-pages, complex reservation systems. OntoBuilder enables fully-automatic ontology matching. The use of ontologies, as opposed to relational schema or XML, as an underlying data model allows a flexible representation of metadata, that can be tailored to many different types of applications. OntoBuilder was developed using Java, which makes it portable to various platforms and operating system environments. We demonstrate OntoBuilder using an easy-to-follow example of matching car rental ontologies. The system creates ontologies of car rental Web sites on-the-fly, and combine them into a global ontology. The benefits of OntoBuilder in resolving, in an automatic manner, semantic heterogeneity, including synonyms and designer errors are highlighted.

Similar to most scientific studies, biological analyses demand a great deal of computations and simulations involving sophisticated tools that are often found geographically distributed over the Internet. A worldwide effort in genomics research has resulted in a powerful collection of publicly available sequence analysis tools. These tools often require specialized local services and domain knowledge to function correctly, rendering them unlikely candidates for integration into remote database applications. Thus, integration of heterogeneous "functions" still remains an open problem. Providing a reasonable framework for seamless integration of these tools with database query engines will enable application developers to exploit and harness the power of these effective analysis tools. In this paper, we present an integration framework for such tools by enabling access to them in a user transparent way as part of database queries. In our system, such online tools are abstracted as remote user defined functions (RUDF). An extended SQL DDL language, called the Internet Function Definition Language (IFDL), is presented for the specification and definition of RUDFs. The interface between database system and the Internet is implemented using a layer based on a language called the Hyper Text Query Language (HTQL). The separation of IFDL, DDL, HTQL and SQL DML offers several optimization opportunities and makes it possible to develop an architecture for interoperability of heterogeneous databases with RUDFs in more simple and efficient ways.

While the idea of extending XML to include object-oriented features has been gaining popularity in general, the potential of inheritance in document design has not been well recognized in contemporary research. In this paper we demonstrate that XML with dynamic inheritance aids better document designs and decreased management overheads and support increased autonomy. As an extended application, we point out that dynamic inheritance also helps effective automated web portal and ontology designs.We present an object-oriented extension to the language of XML to include dynamic inheritance and describe a middle layer that implements our system. We explain our system with several practical examples.

Recent research has focused on computing large item sets for association rule mining using SQL3 least fixpoint computation, and by exploiting the monotonic nature of the SQL3 aggregate functions such as sum and create view recursive constructs. Such approaches allow us to view mining as an ad hoc querying exercise and treat the efficiency issue as an optimization problem. We present a recursive implementation of a recently proposed least fixpoint operator for computing large item sets from object-relational databases. We present experimental evidence to show that our implementation compares well with several well-regarded and contemporary algorithms for large item set generation.

Querying and visualization of phylogenetic databases remain a great challenge due to their complex tree type semi structured nature. Naturally, successful phylogenetic databases such as the Tree of Life database at the University of Arizona are implemented as Web documents in HTML. While Web implementation of such databases facilitate the representation, and in part, visualization of their contents, querying remains an issue. The interoperability of Web-based phylogenetic databases with other similar databases such as TreeBase and RDB which are implemented using traditional database management systems, has not been possible due to the impedance mismatch between the underlying query and data representation framework. In this paper, we present a novel approach to phylogentic database management using existing database technologies without compromising potential opportunities for visualization and interoperability. We present a Web-based tool for the creation, querying and visualization of phylogenetic databases. We demonstrate the functional capabilities and strengths of our system by recreating the Tree of Life database in our system and performing queries that are not possible in the original Tree of Life database.

Although there have been several encouraging attempts at developing methods for data mining using SQL, simplicity and efficiency still remain significant impediments for further development. In this paper, we propose a significantly new approach and show that any object relational database can be mined for association rules without any restructuring or preprocessing using only basic SQL3 constructs and functions, and hence no additional machinery is necessary. In particular, we show that the cost of computing association rules for a given database does not depend on support and confidence thresholds. More precisely, the set of large items can be computed using one simple join query and an aggregation once the set of all possible meets (least fixpoint) of item set patterns in the input table is known. The principal focus of this paper is to demonstrate that several SQL3 expressions exists for the mining of association rules.

No abstract available.