6th ACE Call Awards

Accelerating traditional workflows in animation and gaming through creative AI solutions

PI: Hans-Martin Rall (ADM)
Co-PIs: Seah Hock Soon (SCSE), Ong Yew Soon (SCSE), Davide Benvenuti (ADM), Ng Woon Lam (ADM)
Collaborators: Sascha Robitzky (AI-X, NTU IGP) , Santiago Montesdeoca (Artineering OÜ)


Animation and games remain areas of high economic interest and potential for Singapore in the 21st century. Growth in games sectors is exponential (The Week 2021) along with a substantial increase in demand for motion graphics (artboundinititative.com, 2020). Our project aims to enable the AI-based creation of a wide variety of artistic styles that closely resemble their handmade equivalents. Films like the Oscar-winning Spiderman-Into the Spiderverse (2018) demonstrate the trend towards stylized CG animation that offers graphic variety. Modern computer animation has been able to integrate and mimic organic artforms like charcoal drawing, oil painting, Western watercolour-painting or Chinese painting through new methods in non-photorealistic-rendering (NPR). However, many productions that look for the “human touch” and handmade look for artistic reasons still must rely on laborious frame by frame hand-painted digital retouching. (See e.g. the Academy Award nominated 2017 animated feature film Loving Vincent). A successful solution would provide creative industries with a tool that would minimize need for repetitive labour and enable smaller companies to succeed with fresh ideas on a global scale. By focussing on the recreation of Chinese painting for animation and games, the preservation of advancement of intangible cultural heritage would also be addressed. This adds significant cultural value to the economic potential of the project. Our proposal qualifies for the ACE grant scheme because it has the potential to accelerate workflows in the creative industries with better results through creative AI solutions. This can only be enabled through an art/science collaboration that defies categorization in STEM or Non-STEM fields. Speaking further to that, the project requires the involvement of an AI coding-expert in close collaboration with animation specialists: The resulting EOM costs will exceed the EP4-cap (100 K). The proposal does neither qualify for the existing AI Singapore funding scheme that prioritizes technological expertise over creative exploration.

Untangling cancer re-wiring: Pan-Cancer mapping of transcription factor driven dysregulatory hotspots using AlphaFold2 and integrative machine learning

PI: Li Yinghui (SBS)
Co-PI: Kwoh Chee Keong (SCSE), Mu Yuguang (SBS)

Transcription factors (TFs) are DNA-binding proteins that recognise specific DNA motifs in regulatory elements (REs). Such specificity is defined by their composition and 3D conformation. More than 1600 TFs are thought to be encoded in the genome, each binding thousands of REs throughout the genome. Mutations affecting such REs disrupt healthy TF-driven regulatory networks. Cumulative corruption of these regulatory networks re-wires oncogenes or tumour suppressors potentiating carcinogenesis in any tissue. Despite the development of valuable high-throughput methods to profile TF binding sites such as ChIP-seq, building genome-wide TF binding activity maps for such a large collection of proteins across a diverse range of tissues remains challenging. A recent machine learning breakthrough in protein folding prediction (AlphaFold2) provides a timely opportunity to overcome this hurdle. Effectively, the 3D structures for the entire human proteome including all transcription factors have been accurately predicted. We propose to bring together unique expertise in Cancer Biology, Biophysics and Machine Learning to build an integrative system that can leverage the new 3D structures, DNA accessibility and DNA variation information to predict active TF binding sites for any genome and evaluate their dysregulatory potential. This approach is the first of its kind to utilise 3D TF structures at such a scale to enhance TF binding site discovery throughout the genome. It is also one of the first initiatives to capitalise on the robust AlphaFold2 results released last month. A Convolutional Neural Network approach previously developed to improve Protein-Ligand binding affinity prediction will be adapted for TF binding site discovery. TF binding site predictions will then be integrated with DNA accessibility data to pinpoint tissue specific binding for hundreds of healthy genomes, before applying an attentionbased deep learning model trained on diverse epigenomic contexts to predict the dysregulatory impact of altered TF binding profiles in diseased genomes.

Study of Foxing on Watercolor Paper and its Prevention

PI: Ng Woon Lam (ADM)
Co-PI: Liu Xuewei (SPMS), Gao Yonggui (SBS)


Watercolor is an important traditional art form in Singapore. Though most watercolor art are on archivable artist grade watercolor paper, foxing is an issue. The causes of foxed stains on paper are categorized as biotic and abiotic. Singapore has a large collection of watercolor art in both public (museums, institutions and government sectors) and private collections (banks, corporates and individuals).

Current available preservation methods are costly and only adopted by financially capable museums and collectors. Common prevention requires controlled pH, relative humidity (RH) and room temperature [1] which is hardly feasible to most artists and collectors. This project aims to develop a low-cost, simple and fast preventive method for most artists and collectors. The protection starts earlier and is more cost-effective. It further educates artists that art preservation starts from them instead of the collectors or museums. The developed concepts and methods must be easily understood and readily available for artists to apply.

This research will study the basic types of foxing and their causes on watercolor paper. Based on the literature review of biotic and abiotic foxing and their probable causes, the study will confirm the major forms of foxing stains on artist-grade watercolor paper. Analysis will cover elemental content, organic fingerprints and morphology of the foxed stains, using optical microscopy, FTIR, SEM-EDS and ED-XRF.

After identifying the major types of foxing stains, experiments will be carried out to develop creative protective methods. Similar analytical techniques will be used to study of efficiency of the suggested protective method. Computer simulations will also be used to complement the less feasible prolonged biological culturing and accelerated chemical weather tests.

The research fulfils the ACE direction – a collaborative effort and creative solution from artists and scientists that provide readily available and cost-effective protective measures for watercolors.