Procedural Content Generation via Machine Learning (PCGML) refers to methods that apply machine learning algorithms to generate game content. In particular, the generation of game item descriptions requires techniques to evaluate the similarity between items, and consequently their creativity. This paper improves the BLEU2vec text similarity evaluation technique by integrating it with Byte Pair Encoding (BPE) to capture the relevance of compound words in generated game item descriptions. This novel technique, called Split BLEU2vec, splits compound words into sub-words enabling their similarity evaluation. Our results show that when compared to BLEU2vec baseline, Split BLEu2vec is able to account for semantic embedding of compound words in item descriptions of the game Legend of Zelda.
Co-creative Procedural Content Generation via Machine Learning (PCGML) refers to systems where a PCGML agent and a human work together to produce output content. One of the limitations of co-creative PCGML is that it requires co-creative training data for a PCGML agent to learn to interact with humans. However, acquiring this data is a difficult and time-consuming process. In this work, we propose approximating human-AI interaction data and employing transfer learning to adapt learned co-creative knowledge from one game to a different game. We explore this approach for co-creative Zelda dungeon room generation. CCS CONCEPTS• Computing methodologies → Transfer learning.
e20534 Background: Targeting of the immune inhibitory PD1-PDL1 axis has proven clinically effective in mesothelioma, despite a low somatic mutation burden, and high rate of CDKN2A deletion. Tumour responses to anti PD1 or PDL1 immune checkpoint inhibition are heterogeneous, and the factors underpinning sensitivity remain poorly understood. We therefore addressed this knowledge gap through multi-omic interrogation of tumours from patients enrolled into arm 4 of the Mesothelioma Stratified Therapy umbrella trial (NCT03654833, MiST4), a multi-centre single arm phase IIA trial of atezolizumab and bevacizumab in patients with relapsed mesothelioma. Methods: Next generation sequencing of whole exomes (mesotheliomas and matched germline DNA), transcriptomes, and 16sRNA to profile gut microbiota were undertaken. Spatial phenotyping of the immune landscape employed multiplex immunofluorescence analysis using a 19x depth 4 panel detection. Tumour proportion score (TPS) for PDL1 was assessed using the 22C3 clone, and BAP1, p16ink4a assessed by immunohistochemistry. The MIST4 cohort was dichotomised by best tumour response (50:50) into those patients exhibiting any tumour reduction (R) versus those without (NR). Machine learning (boosting and bagging) was employed to decipher correlates of response. Results: Tumour responses correlated with progression-free survival (PFS, p = 0.0003). Neither PDL1 TPS or CDKN2A expression were predictive. The NR group exhibited a greater degree of genomic instability with higher somatic copy number burden (p = 0.02), homologous recombination deficiency (HRD, p = 0.03), and uniparental disomy (UPD, p = 0.01). Notably the burden of nonsynonymous mutations and neoantigens did not differ significantly between groups. The NR group was transcriptionally enriched for epithelial mesenchymal transition (p < 0.05). Conversely, 16s RNA sequencing revealed higher gut microbial diversity in the R group compared with NR (Shannon index p = 0.009) with R-group enrichment of the type 2 enterotype (provotella 33% vs 9%). R-group enriched genera comprised prevotella, butyricicoccus, bilophilla, Eubacterium ventriosum, whereas the NR group was enriched for erysipelatoclostidium. The log ratio of genera, ie. Log[GR/GNR] was 2-log higher for the R group (p < 0.0001) vs NR group, and was highly predictive of response (with an area under the receiver operator curve of 0.99). Log[GR/GNR] positively correlated with tumour CD8 T cell infiltration (r = 0.6, p = 0.05) and PFS (p = 0.04), but negatively with CD68 monocyte infiltration (p = 0.05), UPD (p = 0.008) and HRD (p = 0.05). Conclusions: We propose a model in which interacting tumour intrinsic and extrinsic factors correlate with response to PDL1-VEGF inhibition in patients with mesothelioma. Gut microbiota composition represents a new, potentially modifiable target with potential to improve immunotherapy outcomes in patients with mesothelioma. Clinical trial information: NCT03654833 .
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.