Luc Dehaspe

Discovery of frequent DATALOG patterns

Discovery of frequent patterns has been studied in a variety of data mining settings. In its simplest form, known from association rule mining, the task is to discover all frequent itemsets, i.e., all combinations of items that are found in a sufficient number of examples. The fundamental task of association rule and frequent set discovery has been extended in various directions, allowing more useful patterns to be discovered with special purpose algorithms. We present WARMR, a general purpose inductive logic programming algorithm that addresses frequent query discovery: a very general DATALOG formulation of the frequent pattern discovery problem.The motivation for this novel approach is twofold. First, exploratory data mining is well supported: WARMR offers the flexibility required to experiment with standard and in particular novel settings not supported by special purpose algorithms. Also, application prototypes based on WARMR can be used as benchmarks in the comparison and evaluation of new special purpose algorithms. Second, the unified representation gives insight to the blurred picture of the frequent pattern discovery domain. Within the DATALOG formulation a number of dimensions appear that relink diverged settings.We demonstrate the frequent query approach and its use on two applications, one in alarm analysis, and one in a chemical toxicology domain.

Clausal Discovery

The clausal discovery engine claudien is presented. CLAUDIEN is an inductive logic programming engine that fits in the descriptive data mining paradigm. CLAUDIEN addresses characteristic induction from interpretations, a task which is related to existing formalisations of induction in logic. In characteristic induction from interpretations, the regularities are represented by clausal theories, and the data using Herbrand interpretations. Because CLAUDIEN uses clausal logic to represent hypotheses, the regularities induced typically involve multiple relations or predicates. CLAUDIEN also employs a novel declarative bias mechanism to define the set of clauses that may appear in a hypothesis.

Mining Association Rules in Multiple Relations

The application of algorithms for efficiently generating association rules is so far restricted to cases where information is put together in a single relation. We describe how this restriction can be overcome through the combination of the available algorithms with standard techniques from the field of inductive logic programming. We present the system Warmr, which extends Apriori [2] to mine association rules in multiple relations. We apply Warmr to the natural language processing task of mining part-of-speech tagging rules in a large corpus of English. be applied to further constrain the space of interesting ARMRs.

Improving the efficiency of inductive logic programming through the use of query packs

Inductive logic programming, or relational learning, is a powerful paradigm for machine learning or data mining. However, in order for ILP to become practically useful, the efficiency of ILP systems must improve substantially. To this end, the notion of a query pack is introduced: it structures sets of similar queries. Furthermore, a mechanism is described for executing such query packs. A complexity analysis shows that considerable efficiency improvements can be achieved through the use of this query pack execution mechanism. This claim is supported by empirical results obtained by incorporating support for query pack execution in two existing learning systems.

A four-gene methylation marker panel as triage test in high-risk human papillomavirus positive patients†

Cervical neoplasia‐specific biomarkers, e.g. DNA methylation markers, with high sensitivity and specificity are urgently needed to improve current population‐based screening on (pre)malignant cervical neoplasia. We aimed to identify new cervical neoplasia‐specific DNA methylation markers and to design and validate a methylation marker panel for triage of high‐risk human papillomavirus (hr‐HPV) positive patients. First, high‐throughput quantitative methylation‐specific PCRs (QMSP) on a novel OpenArray™ platform, representing 424 primers of 213 cancer specific methylated genes, were performed on frozen tissue samples from 84 cervical cancer patients and 106 normal cervices. Second, the top 20 discriminating methylation markers were validated by LightCycler® MSP on frozen tissue from 27 cervical cancer patients and 20 normal cervices and ROCs and test characteristics were assessed. Three new methylation markers were identified (JAM3, EPB41L3 and TERT), which were subsequently combined with C13ORF18 in our four‐gene methylation panel. In a third step, our methylation panel detected in cervical scrapings 94% (70/74) of cervical cancers, while in a fourth step 82% (32/39) cervical intraepithelial neoplasia grade 3 or higher (CIN3+) and 65% (44/68) CIN2+ were detected, with 21% positive cases for ≤CIN1 (16/75). Finally, hypothetical scenario analysis showed that primary hr‐HPV testing combined with our four‐gene methylation panel as a triage test resulted in a higher identification of CIN3 and cervical cancers and a higher percentage of correct referrals compared to hr‐HPV testing in combination with conventional cytology. In conclusion, our four‐gene methylation panel might provide an alternative triage test after primary hr‐HPV testing.

Warmr: a data mining tool for chemical data

Data mining techniques are becoming increasingly important in chemistry as databases become too large to examine manually. Data mining methods from the field of Inductive Logic Programming (ILP) have potential advantages for structural chemical data. In this paper we present Warmr, the first ILP data mining algorithm to be applied to chemoinformatic data. We illustrate the value of Warmr by applying it to a well studied database of chemical compounds tested for carcinogenicity in rodents. Data mining was used to find all frequent substructures in the database, and knowledge of these frequent substructures is shown to add value to the database. One use of the frequent substructures was to convert them into probabilistic prediction rules relating compound description to carcinogenesis. These rules were found to be accurate on test data, and to give some insight into the relationship between structure and activity in carcinogenesis. The substructures were also used to prove that there existed no accurate rule, based purely on atom-bond substructure with less than seven conditions, that could predict carcinogenicity. This results put a lower bound on the complexity of the relationship between chemical structure and carcinogenicity. Only by using a data mining algorithm, and by doing a complete search, is it possible to prove such a result. Finally the frequent substructures were shown to add value by increasing the accuracy of statistical and machine learning programs that were trained to predict chemical carcinogenicity. We conclude that Warmr, and ILP data mining methods generally, are an important new tool for analysing chemical databases.

Presymptomatic Identification of Cancers in Pregnant Women During Noninvasive Prenatal Testing.

IMPORTANCE Noninvasive prenatal testing (NIPT) for fetal aneuploidy by scanning cell-free fetal DNA in maternal plasma is rapidly becoming a major prenatal genetic test. Similar to placental DNA, tumor DNA can be detected in the plasma, and analysis of cell-free tumor DNA can be used to characterize and monitor cancers. We show that plasma DNA profiling allows for presymptomatic detection of tumors in pregnant women undergoing routine NIPT. OBSERVATIONS During NIPT in over 4000 prospective pregnancies by parallel sequencing of maternal plasma cell-free DNA, 3 aberrant genome representation (GR) profiles were observed that could not be attributed to the maternal or fetal genomic constitution. A maternal cancer was suspected, and those 3 patients were referred for whole-body diffusion-weighted magnetic resonance imaging, which uncovered an ovarian carcinoma, a follicular lymphoma, and a Hodgkin lymphoma, each confirmed by subsequent pathologic and genetic investigations. The copy number variations in the subsequent tumor biopsies were concordant with the NIPT plasma GR profiles. CONCLUSIONS AND RELEVANCE We show that maternal plasma cell-free DNA sequencing for noninvasive prenatal testing also may enable accurate presymptomatic detection of maternal tumors and treatment during pregnancy.

Non-invasive detection of genomic imbalances in Hodgkin/Reed-Sternberg cells in early and advanced stage Hodgkin's lymphoma by sequencing of circulating cell-free DNA: a technical proof-of-principle study

BACKGROUND Hodgkins lymphoma is one of the most common lymphoid neoplasms in young adults, but the low abundance of neoplastic Hodgkin/Reed-Sternberg cells in the tumour hampers the elucidation of its pathogenesis, biology, and diversity. After an incidental observation that genomic aberrations known to occur in Hodgkins lymphoma were detectable in circulating cell-free DNA, this study was undertaken to investigate whether circulating cell-free DNA can be informative about genomic imbalances in Hodgkins lymphoma. METHODS We applied massive parallel sequencing to circulating cell-free DNA in a prospective study of patients with biopsy proven nodular sclerosis Hodgkins lymphoma. Genomic imbalances in Hodgkin/Reed-Sternberg cells were investigated by fluorescence in-situ hybridisation (FISH) on tumour specimens. FINDINGS By non-invasive prenatal testing, we observed several genomic imbalances in circulating cell-free DNA of a pregnant woman, who was subsequently diagnosed with early-stage nodular sclerosis Hodgkins lymphoma stage IIA during gestation. FISH on tumour tissue confirmed corresponding genomic imbalances in Hodgkin/Reed-Sternberg cells. We prospectively studied circulating cell-free DNA of nine nodular sclerosis Hodgkins lymphoma cases: eight at first diagnosis and one at first relapse. Seven patients had stage IIA disease and two had stage IVB disease. In eight, genomic imbalances were detected, including, among others, gain of chromosomes 2p and 9p, known to occur in Hodgkins lymphoma. These gains and losses in circulating cell-free DNA were extensively validated by FISH on Hodgkin/Reed-Sternberg cells in biopsy samples. Initiation of chemotherapy induced normalisation of circulating cell-free DNA profiles within 2-6 weeks. The cell cycle indicator Ki67 and cleaved caspase-3 were detected in Hodgkin/Reed-Sternberg cells by immunohistochemistry, suggesting high turnover of Hodgkin/Reed-Sternberg cells. INTERPRETATION In early and advanced stage nodular sclerosis Hodgkins lymphoma, genomic imbalances in Hodgkin/Reed-Sternberg cells can be identified by massive parallel sequencing of circulating cell-free DNA at diagnosis. The rapid normalisation of circulating cell-free DNA profiles on therapy initiation suggests a potential role for circulating cell-free DNA profiling in early response monitoring. This finding creates several new possibilities for exploring the diversity of Hodgkins lymphoma, and has potential implications for the future clinical development of biomarkers and precision therapy for this malignancy. FUNDING KU Leuven-University of Leuven and University Hospitals Leuven.

Noninvasive prenatal testing using a novel analysis pipeline to screen for all autosomal fetal aneuploidies improves pregnancy management.

Noninvasive prenatal testing by massive parallel sequencing of maternal plasma DNA has rapidly been adopted as a mainstream method for detection of fetal trisomy 21, 18 and 13. Despite the relative high accuracy of current NIPT testing, a substantial number of false-positive and false-negative test results remain. Here, we present an analysis pipeline, which addresses some of the technical as well as the biologically derived causes of error. Most importantly, it differentiates high z-scores due to fetal trisomies from those due to local maternal CNVs causing false positives. This pipeline was retrospectively validated for trisomy 18 and 21 detection on 296 samples demonstrating a sensitivity and specificity of 100%, and applied prospectively to 1350 pregnant women in the clinical diagnostic setting with a result reported in 99.9% of cases. In addition, values indicative for trisomy were observed two times for chromosome 7 and once each for chromosomes 15 and 16, and once for a segmental trisomy 18. Two of the trisomies were confirmed to be mosaic, one of which contained a uniparental disomy cell line. As placental trisomies pose a risk for low-grade fetal mosaicism as well as uniparental disomy, genome-wide noninvasive aneuploidy detection is improving prenatal management.

Maximum Entropy Modeling with Clausal Constraints

We present the learning system Maccent which addresses the novel task of stochastic MAximum ENTropy modeling with Clausal Constraints. Maximum Entropy method is a Bayesian method based on the principle that the target stochastic model should be as uniform as possible, subject to known constraints. Maccent incorporates clausal constraints that are based on the evaluation of Prolog clauses in examples represented as Prolog programs. We build on an existing maximum-likelihood approach to maximum entropy modeling, which we upgrade along two dimensions: (1) Maccent can handle larger search spaces, due to a partial ordering defined on the space of clausal constraints, and (2) uses a richer first-order logic format. In comparison with other inductive logic programming systems, MACCENT seems to be the first that explicitly constructs a conditional probability distribution p(C|I) based on an empirical distribution \(\tilde p\)(C|I) (where p(C|I) (\(\tilde p\)(C|I)) equals the induced (observed) probability of an instance I belonging to a class C). First experiments indicate MACCENT may be useful for prediction, and for classification in cases where the induced model should be combined with other stochastic information sources.