Learning state machine-based string edit kernels

During the past few years, several works have been done to derive string kernels from probability distributions. For instance, the Fisher kernel uses a generative model M (e.g. a hidden Markov model) and compares two strings according to how they are generated by M. On the other hand, the marginalized kernels allow the computation of the joint similarity between two instances by summing conditional probabilities. In this paper, we adapt this approach to edit distance-based conditional distributions and we present a way to learn a new string edit kernel. We show that the practical computation of such a kernel between two strings x and x^′ built from an alphabet Σ requires (i) to learn edit probabilities in the form of the parameters of a stochastic state machine and (ii) to calculate an infinite sum over Σ^* by resorting to the intersection of probabilistic automata as done for rational kernels. We show on a handwritten character recognition task that our new kernel outperforms not only the state of the art string kernels and string edit kernels but also the standard edit distance used by a neighborhood-based classifier.     

Good edit similarity learning by loss minimization

Similarity functions are a fundamental component of many learning algorithms. When dealing with string or tree-structured data, measures based on the edit distance are widely used, and there exist a few methods for learning them from data. However, these methods offer no theoretical guarantee as to the generalization ability and discriminative power of the learned similarities. In this paper, we propose an approach to edit similarity learning based on loss minimization, called GESL. It is driven by the notion of (?,??,??)-goodness, a theory that bridges the gap between the properties of a similarity function and its performance in classification. Using the notion of uniform stability, we derive generalization guarantees that hold for a large class of loss functions. We also provide experimental results on two real-world datasets which show that edit similarities learned with GESL induce more accurate and sparser classifiers than other (standard or learned) edit similarities.     

Time to relapse has prognostic value in patients with aggressive lymphoma enrolled onto the Parma trial

PURPOSE: The purpose of this study was to investigate the prognostic value of time to relapse in 188 adult patients with intermediate- or high-grade non-Hodgkin's lymphoma (NHL) included on the Parma trial at the time of their first relapse. PATIENTS AND METHODS: The median follow-up of these patients is 102 months after registration onto the Parma study. Time to relapse was calculated from initial diagnosis, and a cutoff of 12 months was used to separate 77 patients defined as early relapse from 111 patients defined as late relapse. RESULTS: Patients with early and late relapses had significantly different overall response rates to salvage therapy with two courses of dexamethasone, high-dose cytarabine, and cisplatin (DHAP; 40% v 69%; P=.00007) and different 8-year survival rates (13% v 29%; P=.00001). Features at relapse with a negative prognostic value in univariate analysis were higher than normal lactic dehydrogenase (LDH) levels, tumor size greater than 5 cm, Ann Arbor stages III to IV, and Karnofsky score less than 80%. Therefore, multivariate analyses were performed. Time to relapse (P=.001) and LDH levels at relapse (P=.003) had independent prognostic value, whereas tumor size did not reach statistical significance in the logistic model that predicted overall response after two courses of DHAP. The study of prognostic factors for overall survival (OS) and progression-free survival (PFS) confirmed the prognostic value of time to relapse (P < .0001 for OS and P=.005 for PFS) independent of response or treatment after two courses of DHAP. CONCLUSION: Time to relapse may be used to stratify patients at time of first relapse of intermediate to high-grade non-Hodgkin's lymphoma.     

Ageing of EPDM elastomers exposed to γ-radiation studied by H broadband and C high-resolution solid-state NMR

Ageing of EPDM synthetic elastomers exposed to γ-irradiation has been investigated using solid-state NMR Spectroscopy. Both ¹³C high-resolution and ¹H wideline measurements were carried out to evaluate chemical degradation including oxidation products formation, chain scission and crosslinking phenomena. Highly specific structures were tentatively proposed as a result of careful examination of ¹³C chemical shifts. The importance of diene monomer, the stabilising effect of curing process and the addition of antioxidants were evaluated on the basis of several polymer compositions.     

Observation by soft X-ray laser interferometry of Nb cathodesurface evolution during very high electric field application

We present the very first experiment allowing in situ observation of surface evolution during very high electric field application. This has been achieved by soft (λ=21.2 nm) X-ray laser Fresnel two-mirror interferometry. The surface under study was a niobium planar cathode opposite to a stainless steel blade-like anode. The applied field range on the cathode was F≈0 to 50 MV/m. The interferograms were single shots of 80 ps duration, allowing to probe the surface morphology quasi-instantaneously. The delay between two shots was 20 min while the electric field was kept on. The comparison of successive interferograms gives the Nb surface evolution over ~2 h. Thus, we observed the appearance of defects of ~3 nm high and their evolution with time. These may be the precursors of the minute breakdown which occurred later on. This technique thus makes it possible to detect the preliminary stages of a discharge     

Mutations in the environment of the primary quinone facilitate proton delivery to the secondary quinone in bacterial photosynthetic reaction centers

In Rhodobacter capsulatus, we constructed a quadruple mutant that reversed a structural asymmetry that contributes to the functional asymmetry of the two quinone sites. In the photosynthetically incompetent quadruple mutant RQ, two acidic residues near QB, L212Glu and L213Asp, have been mutated to Ala; conversely, in the QA pocket, the symmetry-related residues M246Ala and M247Ala have been mutated to Glu and Asp. We have selected photocompetent phenotypic revertants (designated RQrev3 and RQrev4) that carry compensatory mutations in both the QA and QB pockets. Near QA, the M246Ala --> Glu mutation remains in both revertants, but M247Asp is replaced by Tyr in RQrev3 and by Ala in RQrev4. The engineered L212Ala and L213Ala substitutions remain in the QB site of both revertants but are accompanied by an additional electrostatic-type mutation. To probe the respective influences of the mutations occurring near the QA and QB sites on electron and proton transfer, we have constructed two additional types of strains. First, "half" revertants were constructed that couple the QB site of the revertants with a wild-type QA site. Second, the QA sites of the two revertants were linked with the L212Glu-L213Asp --> Ala-Ala mutations of the QB site. We have studied the electron and proton-transfer kinetics on the first and second flashes in reaction centers from these strains by flash-induced absorption spectroscopy. Our data demonstrate that substantial improvements of the proton-transfer capabilities occur in the strains carrying the M246Ala --> Glu + M247Ala --> Tyr mutations near QA. Interestingly, this is not observed when only the M246Ala --> Glu mutation is present in the QA pocket. We suggest that the M247Ala --> Tyr mutation in the QA pocket, or possibly the coupled M246Ala --> Glu + M247Ala --> Tyr mutations, accelerates the uptake and delivery of protons to the QB anions. The M247Tyr substitution may enable additional pathways for proton transfer that are located near QA.     

S 16924 ((R)-2-[1-[2-(2,3-dihydro-benzo[1,4] dioxin-5-yloxy)-ethyl]-pyrrolidin-3yl]-1-(4-fluoro-phenyl)-ethanone), a novel, potential antipsychotic with marked serotonin (5-HT)1A agonist properties: II. Functional profile in comparison to clozapine and haloperidol

S 16924 antagonized locomotion provoked by dizocilpine and cocaine, reduced conditioned avoidance responses and blocked climbing elicited by apomorphine, models predictive of control of the positive symptoms of schizophrenia: its median inhibitory dose (ID)50 was 0.96 mg/kg, s.c. vs. 1.91 for clozapine and 0.05 for haloperidol. Rotation elicited in unilateral, substantia nigra-lesioned rats by the D1 agonist, SKF 38393, and by the D2 agonist, quinpirole, was blocked equipotently by S 16924 (0.8 and 1. 7) and clozapine (0.6 and 2.0), whereas haloperidol preferentially blocked quinpirole (0.02) vs. SKF 38393 (1.8). S 16924 more potently inhibited the head-twitches elicited by 1-(2, 5-dimethoxy-4-iodophenyl)-2-aminopropane (DOI) and the locomotion provoked by phencyclidine than it inhibited the locomotion elicited by amphetamine (ID50s = 0.15 and 0.02 vs. 2.4). Clozapine showed a similar preference (0.04 and 0.07 vs. 8.6), but not haloperidol (0. 07 and 0.08 vs. 0.04). The discriminative stimulus (DS) properties of DOI were also blocked by S 16924 (ID50 = 0.17) and clozapine (0. 05) but not by haloperidol (>0.16). S 16924 fully (100%) generalized [effective dose (ED)50 = 0.7] to a clozapine DS and clozapine (0.23) fully generalized to a S 16924 DS whereas haloperidol (>/=0.08) only partially generalized (/=80.0) or clozapine (>/=80.0). Further, S 16924 (ID50 = 3.2) and clozapine (5.5) inhibited induction of catalepsy by haloperidol. This action of S 16924 was abolished by the 5-HT1A receptor antagonist, WAY 100,635 (0.16), which less markedly attenuated the anticataleptic action of clozapine. Further, although gnawing elicited by methylphenidate was inhibited by S 16924 (ID50 = 8.4), clozapine (19.6) and haloperidol (0.04), only the action of S 16924 was blocked by WAY 100,635 (0.16). Haloperidol potently (0.01-0.16, approximately 24-fold) increased prolactin levels whereas they were less markedly affected by S 16924 (2.5-40.0, 4-fold) and clozapine (10.0-40.0, 3-fold). Clozapine displayed high affinity at cloned, human, muscarinic (M1) and native, histamine (H1) receptors (Kis = 4.6 and 5.4 nM, respectively), whereas S 16924 (>1000 and 158) and haloperidol (>1000 and 453) displayed low affinity. In conclusion, S 16924 displays a profile of activity in diverse models of potential antipsychotic and extrapyramidal properties similar to that of clozapine and different to that of haloperidol. In particular, reflecting its partial agonist actions at 5-HT1A receptors, S 16924 inhibits rather than induces catalepsy in rats. However, in contrast to clozapine, S 16924 displays only low affinity for muscarinic and histaminic receptors.