Search

Tämä strategisen tutkimuksen neuvoston rahoittaman Kestävän kasvun avaimet (GROWTH) -ohjelman julkaisu tuottaa tutkimukseen perustuvia näkökulmia ja ratkaisun avaimia päätöksentekijöiden hyödynnettäväksi tukemaan oikeudenmukaista siirtymää kohti kestävää ja hyvinvoivaa yhteiskuntaa. Nämä ratkaisun avaimet korostavat uudentyyppistä arvonluontia, korkeatasoista koulutusta sekä sukupolvien välisen oikeudenmukaisuuden toteutumista kestävyysmurroksessa.

There’s been a significant growth in the gaming industry, which has lead to an increased number of collected player and usage data, including game events, player interactions, the connections between players and individual preferences. Such big data has many use cases such as the identification of gaming bottlenecks, detection and prediction of anomalies and suspicious usage patterns for security, and real time offer specification via fine-grained user profiling based on their interest profiles. Offering personalized offer timing could reduce product cannibalization, and ethical methods increase the trust of customers. The goal of this thesis is to predict the value and time of the next in-game purchase in a mobile game. Using data aggregation, event-based purchase data, daily in-game behaviour metrics and session data are combined into a single data table, from which samples of 50 000 data points are taken. The features are analyzed for linear correlation with the labels, and their combinations are used as input for three machine learning algorithms: Random Forest, Support Vector Machine and Multi-Layer Perceptron. Both purchase value and purchase time are correlated with features related to previous purchase behaviour. Multi-Layer Perceptron showed the lowest error in predicting both labels, showing an improvement of 22,0% for value in USD and 20,7% for days until purchase compared to a trivial baseline predictor. For ethical customer behaviour prediction, sharing of research knowledge and customer involvement in the data analysis process is suggested to build awareness.Progame : tapahtumapohjainen koneoppimisjärjestelmä pelinsisäiseen markkinointiin. Tiivistelmä. Peliteollisuuden kasvu on johtanut kerättävän pelaaja- ja käyttödatan määrään nousuun, koostuen mm. pelitapahtumista, interaktiodatasta, pelaajien välisistä yhteyksistä ja henkilökohtaisista mieltymyksistä. Tällaisella massadatalla on monia käyttötarkoituksia kuten tietoliikenteen teknisten rajoitusten tunnistaminen pelikäytössä, käyttäjien tavallisuudesta poikkeavan käytöksen tunnistaminen ja ennustaminen tietoturvatarkoituksiin, sekä reaaliaikainen tarjousten määrittäminen hienovaraisella käyttäjien mieltymysten profiloinnilla. Ostotarjousten henkilökohtaistaminen voi vähentää uusien tuotteiden aiheuttamaa vanhojen tuotteiden myynnin laskua, ja eettiset menetelmät parantavat asiakkaiden luottamusta. Tässä työssä ennustetaan asiakkaan seuraavan pelinsisäisen oston arvoa ja aikaa mobiilipelissä. Tapahtumapohjainen ostodata, päivittäiset pelin sisäiset metriikat ja sessiodata yhdistetään yhdeksi datataulukoksi, josta otetaan kerrallaan 50 000:n datarivin näytteitä. Jokaisen selittävän muuttujan lineaarinen korrelaatio ennustettavan muuttujan kanssa analysoidaan, ja niiden yhdistelmiä käytetään syötteenä kolmelle eri koneoppimismallille: satunnainen metsä (Random Forest), tukivektorikone (Support Vector Machine) ja monikerroksinen perseptroniverkko (Multi-Layer Perceptron). Tutkimuksessa havaittiin, että sekä tulevan oston arvo että ajankohta korreloivat aiemman ostokäyttäytymisen kanssa. Monikerroksisella perseptroniverkolla oli pienin virhe molemmille ennustettaville muuttujille, ja verrattuna triviaaliin vertailuennustimeen, se vähensi virhettä 22,0% arvon ennustamisessa ja 20,7% seuraavaan ostoon jäljellä olevien päivien ennustamisessa. Eettisen asiakkaiden käyttäytymisen ennustamisen varmistamiseksi ja tietoisuuden lisäämiseksi ehdotetaan tutkimustiedon jakamista ja asiakkaan ottamista mukaan analyysin tekemiseen.

The importance of intrinsically disordered late embryogenesis abundant (LEA) proteins in the tolerance to abiotic stresses involving cellular dehydration is undisputed. While structural transitions of LEA proteins in response to changes in water availability are commonly observed and several molecular functions have been suggested, a systematic, comprehensive and comparative study of possible underlying sequence-structure-function relationships is still lacking. We performed molecular dynamics (MD) simulations as well as spectroscopic and light scattering experiments to characterize six members of two distinct, lowly homologous clades of LEA_4 family proteins from Arabidopsis thaliana. We compared structural and functional characteristics to elucidate to what degree structure and function are encoded in LEA protein sequences and complemented these findings with physicochemical properties identified in a systematic bioinformatics study of the entire Arabidopsis thaliana LEA_4 family. Our results demonstrate that although the six experimentally characterized LEA_4 proteins have similar structural and functional characteristics, differences concerning their folding propensity and membrane stabilization capacity during a freeze/thaw cycle are obvious. These differences cannot be easily attributed to sequence conservation, simple physicochemical characteristics or the abundance of sequence motifs. Moreover, the folding propensity does not appear to be correlated with membrane stabilization capacity. Therefore, the refinement of LEA_4 structural and functional properties is likely encoded in specific patterns of their physicochemical characteristics.

The targeted pathogen-selective approach to drug development holds promise to minimize collateral damage to the beneficial microbiome. The AB₅-topology pertussis toxin (PtxS1-S5) is a major virulence factor of Bordetella pertussis, the causative agent of the highly contagious respiratory disease whooping cough. Once internalized into the host cell, PtxS1 ADP-ribosylates α-subunits of the heterotrimeric Gαi-superfamily, thereby disrupting G-protein-coupled receptor signaling. Here, we report the discovery of the first small molecules inhibiting the ADP-ribosyltransferase activity of pertussis toxin. We developed protocols to purify milligram-levels of active recombinant B. pertussis PtxS1 from Escherichia coli and an in vitro high throughput-compatible assay to quantify NAD⁺ consumption during PtxS1-catalyzed ADP-ribosylation of Gαi. Two inhibitory compounds (NSC228155 and NSC29193) with low micromolar IC₅₀-values (3.0 μM and 6.8 μM) were identified in the in vitro NAD⁺ consumption assay that also were potent in an independent in vitro assay monitoring conjugation of ADP-ribose to Gαi. Docking and molecular dynamics simulations identified plausible binding poses of NSC228155 and in particular of NSC29193, most likely owing to the rigidity of the latter ligand, at the NAD⁺-binding pocket of PtxS1. NSC228155 inhibited the pertussis AB₅ holotoxin-catalyzed ADP-ribosylation of Gαi in living human cells with a low micromolar IC₅₀-value (2.4 μM). NSC228155 and NSC29193 might prove to be useful hit compounds in targeted B. pertussis-selective drug development.

Acute kidney injury (AKI) is a syndrome that frequently affects the critically ill. Recently, an increased number of dinucleotide repeats in the HMOX1 gene were reported to associate with development of AKI in cardiac surgery. We aimed to test the replicability of this finding in a Finnish cohort of critically ill septic patients. This multicenter study was part of the national FINNAKI study. We genotyped 300 patients with severe AKI (KDIGO 2 or 3) and 353 controls without AKI (KDIGO 0) for the guanine–thymine (GTn) repeat in the promoter region of the HMOX1 gene. The allele calling was based on the number of repeats, the cut off being 27 repeats in the S–L (short to long) classification, and 27 and 34 repeats for the S–M–L₂ (short to medium to very long) classification. The plasma concentrations of heme oxygenase-1 (HO-1) enzyme were measured on admission. The allele distribution in our patients was similar to that published previously, with peaks at 23 and 30 repeats. The S-allele increases AKI risk. An adjusted OR was 1.30 for each S-allele in an additive genetic model (95% CI 1.01–1.66; p = 0.041). Alleles with a repeat number greater than 34 were significantly associated with lower HO-1 concentration (p