The best approach would be starting with smaller projects involving neural networks and genetic algorithms to gain experience in order to speedup the coding of the project you have proposed; playing around with TensorFlow and Unreal Engine it is not a bad idea.
Hint: when implementing your idea of artificial life, you should consider that each cell/organism have to have some kind of sensors in order to capture informations from the environment; such informations i.e. the position and the distance of the nearest meal and/or predators, the temperature, the pressure and depth of water, should be passed through the neural network to determine the response of the cell. Also, in your environment you should promote the spreading of organisms which responses are euristically better i.e. cells that don't get caught by predators or don't die by starvation. How? Simply by evolving their brain/brains/sensors through a genetic algorithm that favors individuals/species with good parameters. I recommend you a nature-inspired AI method, it is called NEAT model. It explains how to implement a neural networks that can be evolved. The paper can be found here: Evolving Neural Networks through Augmenting Topologies.
A different approach to NEAT would be Deep Reinforcement Learning; in the link you can find a demo artifical organism that learns how to find meals.
There are a ton of parameters and implementations you can consider, the only limit is your creativity.